Estimating Air Travel
Demand Elasticities
Final Report
strategic
transportation
& tourism
solutions
Prepared
Prepared for
for
IATA
IATA
Prepared
Prepared by
by
InterVISTAS
Consulting
InterVISTAS Consulting Inc.
Inc.
Draft of December 2006
28 December 2007
Estimating Air Travel Demand Elasticities
Page i
Executive Summary
This report summarises analysis which examines fare elasticities in the passenger aviation market
– the demand response by air passengers to fare increases or decreases. The aim of the study is
to provide robust elasticity estimates to address policy issues related to liberalisation, airport
charges, taxation, emissions schemes, etc. The key study components were:
Extensive Literature Review
A literature review was conducted to examine previous research on air fare elasticities over the last
25 years. Over 23 papers were reviewed, from the following conclusions were developed:
ƒ
ƒ
ƒ
ƒ
ƒ
Sensitivity to Air Fare Changes
All of the studies reviewed, spanning a period of over 25 years, found that there was a
significant demand response to changes in air fares, such that increases in air fare lead to
lower passenger traffic demand. The consistency of this result strongly indicates that any
policy action that results in higher fares (e.g., taxes, increased landing fees) will result in a
decline in demand. The actual decline in demand will depend on a number of factors, as
discussed below.
Business Versus Leisure Passengers
In general, the results show that, all else being equal, business travellers are less sensitive to
fare changes (less elastic) than leisure travellers. Intuitively, this result is plausible – business
travellers generally have less flexibility to postpone or cancel their travel than leisure travellers.
Nevertheless, the studies do show that even business travel will decline in the face of fare
increases, albeit not to the say expect as leisure travel.
Short-Haul Versus Long-Haul Travel
Another fairly consistent finding was that fare elasticities on short-haul routes were generally
higher than on long-haul routes. In par, this reflects the opportunity for inter-modal substitution
on short haul routes (e.g., travellers can switch to rail or car in response to air fare increases).
Carrier Versus Market Versus National Elasticities
Some of the studies supported the idea that the demand elasticity faced by individual air
carriers is higher than that faced by the whole market. For example, Oum, Zhang, and Zhang
(1993) estimated firm-specific elasticities in the U.S. and estimated values ranging from -1.24
to -2.34, while studies estimating market or route elasticities ranged from -0.6 to -1.8. In
contrast, Alperovich and Machnes (1994) and Njegovan (2006) used national-level measures
of air travel in Israel and the UK respectively and produced even lower elasticity values (-0.27
and -0.7, respectively).
Income Elasticities
Many of the studies also including an income variable, in part, to isolate the effects of a shift
along the demand curve (such as would be caused by a price change) from the effect of a shift
in the demand curve (as might be caused by a change in income levels). The studies including
the income term all produced positive income elasticities, as would be expected (air travel
increases as incomes increase). Virtually all of these studies estimated income elasticities
above one, generally between +1 and +2. This indicates air travel increases at a higher rate
than incomes.
28 December 2007
Estimating Air Travel Demand Elasticities
Page ii
Econometric Analysis of a Various Datasets to Estimate Additional Elasticities
Econometric analysis was conducted on three different datasets:
ƒ
ƒ
ƒ
U.S. DB1A data (U.S. domestic). A U.S. government database with a 10% random sampling
of all tickets purchased in the U.S. for travel on U.S. airlines going back to 1993. Quarterly
traffic and fare data was taken for the period from 1994 Q1 to 2005 Q4 for the top 1000 city
pair routes (by traffic).
PaxIS (worldwide traffic). This database captures market data through IATA’s Billing and
Settlement Plan (BSP) and uses statistical estimates to address missing direct sales, low cost
carriers, charter flight operators, under-represented BSP markets, and non-BSP markets. The
advantage of data series is that traffic and fare estimates are available for routes around the
world. However, this dataset is limited by the short time series (data is available from 2005).
International Passenger Survey (UK outbound traffic). The International Passenger Survey
(IPS) is a survey of a random sample of passengers entering or leaving the UK by air, sea or
the Channel Tunnel. This report exclusively used outbound leisure air passenger traffic data
from the IPS. Quarterly traffic and fare data was taken for the period from 2003 Q2 to 2006
Q2.
Over 500 regression models were estimated. The key results are provided in the full report.
General Guidelines on Air Fare Elasticity Values
The literature review and econometric analysis demonstrated that air fares elasticities vary
depending on a number of factors such as geography, distance and level of aggregation.
Ultimately, the determining the right elasticity value to use depends on the type of question being
asked. Examining the traffic impact of a fare increase on a given route requires a different
elasticity than when examining the impact of an across-the-board fare increase on all routes in a
country or region.
In this study, the results of the literature review and econometric analysis were synthesised to
develop some general guidelines on the use and application of air fare elasticities. While the
literature review and econometrics did not provide a full set of elasticities covering all possibilities,
they did provide sufficient information to infer a full set of elasticities.
The following elasticities have been developed as a synthesis of the literature review and the
econometric analysis. The approach taken was to develop three base elasticities reflecting the
levels of aggregation (route, national and pan-national level). Multiplicative adjustors were then
developed to adjust the elasticities to reflect specific markets.
Base Elasticities
The base elasticities were developed from the analysis of the various datasets and the generalised
findings from the literature review.
ƒ
Route/Market Level: -1.4
The literature review found that elasticities at the route or market level in the range of -1.2 to 1.5. This was verified by our own econometric analysis of the U.S. DB1A where it was
possible to capture the effects of route substitution through the use of route dummy variables
and experiments using variables capturing the price of route substitutes. These regressions
28 December 2007
Estimating Air Travel Demand Elasticities
ƒ
ƒ
Page iii
produced fare elasticities in the region of -1.4, which has been used as the base elasticity at
the route level.
National Level: - 0.8
The econometric analysis of all three datasets found that without the route substitution term,
the analysis produced elasticities in the region of -0.8. This elasticity is essentially a
combination of the route own price elasticity with cross price elasticities when all national
routes have prices which vary identically. Thus, the less elastic result is consistent with
observations that part of the so called price elasticity observed from LCCs at secondary
airports involves diversion from primary airports in the catchment area, or from diversion from
trips on other routes. When this is controlled for, LCCs have a lower level of market
stimulation, consistent with less elastic national elasticities.
Pan-National Level: -0.6
This result is conceptually derived from the same mathematics which show the relationship
between route and national elasticities. In this case, a much broader set of routes (e.g., across
a continent) experience an identical price change.
Thus the route elasticity is applicable to a situation where the price of an individual route changes.
For example, the fare on Warsaw-Coventry increases but the price of routes from Warsaw to other
UK and other European points remain unchanged. The national elasticity applies to situations
such as all Warsaw-UK prices changing identically, but the price from Warsaw to other European
points being unchanged. Pan-national changes apply where prices from Warsaw to all points in
Europe change identically.
Geographic Aviation Market
The econometric analysis of the IATA PaxIS data found considerable differences between aviation
markets. Based on this analysis, the following elasticity multipliers have been developed:
Geographic Market
Elasticity Multiplier
Intra North America
1.00
Most research uses US data. This is our
reference point.
1.40
Shorter average distances, observed use of very
low fares resulting great market stimulation. The
significantly low fares in Europe (relative to North
America) are consistent with higher elasticities in
Europe. Traditionally the European market had
high charter carrier share, which today is merely
being converted to very low fare LCCs.
Intra Asia
0.95
The LCC phenomena is emerging in Asia, but
modest sized middle class in many markets
suggests somewhat less elastic than in North
America
Intra Sub-Sahara Africa
0.60
These economies have limited middle class,
resulting in high weight on higher income
individuals who are less elastic
Intra South America
1.25
There is an emerging middle class which makes
the market more elastic than sub-Saraha Africa,
Intra Europe
28 December 2007
Comment
Estimating Air Travel Demand Elasticities
Geographic Market
Page iv
Elasticity Multiplier
Comment
and LCCs are emerging in Brazil, Chile, and
Mexico.
1.20
This market is often observed to have fares only
slightly higher than domestic U.S. fares,
consistent with high price elasticity. Market has
been well developed by charter carriers,
consistent with high price elasticity. Price is
likely more important than frequency in this
market than in domestic U.S.
Trans Pacific
(North America – Asia)
0.60
TransPacific has had no charter services, and
continues to have major markets (Japan, China)
with less liberal pricing provisions. Some
emergence of long haul LCCs (e.g., Oasis) but at
present this market seems to be less elastic than
domestic US and than the well developed trans
Atlantic which serves a substantial middle class
Europe – Asia
0.90
This market has marginally lower elasticities than
the U.S. domestic market.
Trans Atlantic
(North America – Europe)
Short-Haul/Long-Haul Adjustment
The literature review consistently found that the fare elasticities on short-haul routes were generally
higher than on long-haul routes. In part, this reflects the opportunity for inter-modal substitution on
short haul routes (e.g., travellers can switch to rail or car in response to air fare increases). While
the geographic breakdowns capture some variation by length of haul, there is still considerable
variation within each market. In particular, very short-haul flights (approximately less than 1 hour
flight time) are subject to greater competition from other modes.
On this basis, the following short-haul multiplicative adjustors can be applied to analysis of shorthaul routes:
ƒ
Short haul: 1.1 (+10%).
Note that this adjustor does not apply to analysis of the trans-Atlantic and trans-Pacific market,
which are considered entirely long haul, with virtually no opportunity for modal substitution.
Application of Adjustors
Elasticities for different situations can be developed by selecting the relevant base elasticity and
applying the relevant multipliers, for example:
1.
To examine the impact of an EU-wide aviation tax on short-haul markets, the elasticity
would be developed as follows:
- Base multiplier: -0.6 (pan-national)
- Geographic market: 1.4 (Intra Europe)
- Short-haul adjustor: 1.1
The multiplier would then be calculated as: -0.6 x 1.4 x 1.1 = -0.924.
28 December 2007
Estimating Air Travel Demand Elasticities
2.
Page v
To examine the impact of a UK tax on aviation on Trans Atlantic traffic, the elasticity
should be developed as follows:
- Base multiplier: -0.8 (national)
- Geographic market: 1.2 (Trans Atlantic)
The multiplier would then be calculated as: -0.8 x 1.2 = -0.96.
3.
To examine the impact of impact of an increase in airport landing fees on a particular
short-haul route in Asia, the elasticity should be developed as follows:
- Base multiplier: -1.4 (route)
- Geographic market: 0.95 (Intra Asia)
- Short-haul adjustor: 1.1
The multiplier would then be calculated as: -1.4 x 0.95 x 1.1 = -1.46.
The full range of possible elasticities is presented in the table below. The route level elasticities
range from -0.84 to -1.96 depending on the geographic market and length of haul. The national
level elasticities range from -0.48 to -1.23, while the pan-national elasticities range from -0.36 to 0.92.
Route/Market Level
National Level
Pan-National Level
Shorthaul
Longhaul
Shorthaul
Longhaul
Shorthaul
Longhaul
Intra North America
-1.54
-1.40
-0.88
-0.80
-0.66
-0.60
Intra Europe
-1.96*
-1.96
-1.23
-1.12
-0.92
-0.84
Intra Asia
-1.46
-1.33
-0.84
-0.76
-0.63
-0.57
Intra Sub-Sahara Africa
-0.92
-0.84
-0.53
-0.48
-0.40
-0.36
Intra South America
-1.93
-1.75
-1.10
-1.00
-0.83
-0.75
Trans Atlantic (North America – Europe)
-1.85
-1.68
-1.06
-0.96
-0.79
-0.72
Trans Pacific (North America – Asia)
-0.92
-0.84
-0.53
-0.48
-0.40
-0.36
Europe-Asia
-1.39
-1.26
-0.79
-0.72
-0.59
-0.54
*The short-haul adjustor has not been applied to the Intra Europe short-haul elasticity in order to maintain elasticities below 2.0
28 December 2007
Estimating Air Travel Demand Elasticities
Page vi
Conclusions – Reconciling the Findings
As described in the literature review, the consensus among most aviation economists has been
that demand for airline services is generally both price and income elastic. The latter suggests that
air transport has the characteristics of a luxury good.1 A natural question is how an income elastic
good can also be price elastic.
This report has revealed that there are different price elasticities associated with different uses.
When consumers are choosing between airlines on a route, or even between destinations for
holidays, conference locations, etc., there is a degree of price elasticity for airline seats. However,
if all competitors on a route, or if a wide range of routes all experience the same proportionate price
increase, the demand for airline services becomes less elastic. As a price increase is extended to
ever larger groups of competing airlines or competing destinations, then the overall demand for air
travel is revealed to be somewhat inelastic.
The implications of this are as follows:
ƒ
ƒ
ƒ
For an airline on a given route, increasing price is likely to result in a more than proportionate
decrease in air travel. Lower fares will greatly stimulate traffic and raise revenues.
(i.e., airline specific fare changes are price elastic).
If all airlines on a given route increase fares by the same amount (e.g., due to the imposition of
passenger based airport fees that are passed on to the consumer), then the decrease in traffic
will be less than the former case.
If all airlines on a wide set of routes increase fares by roughly similar amounts (e.g., due to the
imposition of new market-wide taxes or to the working through of higher fuel or security costs)
then the decrease in traffic may be less or much less than proportional to the increase in fares.
(i.e., general increases in airline fares across a broad range of markets appear to be price
inelastic.)
Thus, the particular elasticity value to be used for analysing price effects in airline markets depends
on the question being asked. The narrower the applicability of a price change, the more elastic the
response. The more general the applicability of a price change (perhaps due to higher costs or
taxes) the less elastic the response.
1
“Luxury good” is a term of art in economics that merely indicates that demand increases more than proportionately with income.
Education, medical services, entertainment and other goods fit this category. The vernacular use of the term “luxury” may have
value judgements attached, which are not intentional in the context of this report.
28 December 2007
Estimating Air Travel Demand Elasticities
Page vii
Glossary of Terms/Abbreviations
Autoregressive Distributed Lag: A regression technique using lagged values of the dependent
and/or independent variable(s)
ARDL: Autoregressive Distributed Lag
BSP: IATA’s Billing and Settlement Plan.
CAA: The United Kingdom Civil Aviation Authority
Consumer Price Index (CPI): A measure of inflation within a given country using consumer prices
of a basket of goods
Cross Price Elasticity: The sensitivity of demand for a particular good to changes in the price of
another good
DB1B: U.S. Department of Transport Database 1B. One of three databases used in this study.
Demand Elasticity: See Price Elasticity
Dummy Variable: An explanatory variable in regression analysis used to capture qualitative
factors. Assumes the value of ‘1’ when the factor is observed and ‘0’ otherwise
Fare Elasticity: See Price Elasticity
Goodness of Fit: A measure of the explanatory variables’ ability in regression analysis to explain
changes in the dependent variable. Assumes a value between 0 and 1, with 1 denoting a perfectly
explained relationship and 0 reflecting no relationship.
Gross domestic product (GDP): A measure of the money value of final goods and services
produced as a result of economic activity in the nation.
Income Elasticity: The sensitivity of demand for a good to changes in income.
IPS: International Passenger Survey. One of three databases used in this study.
LCC: Low Cost Carrier
Luxury Good: A good with an income elasticity above one.
O/D: Origin/Destination
Ordinary Least Squares (OLS): A regression technique that minimizes the sum of squared
residuals from the explanatory variables.
PaxIS: IATA passenger intelligence services. One of three databases used in this study.
Population Elasticity: The sensitivity of demand for a good to changes in population size.
Price Elasticity: Consumers’ sensitivity to price changes for a particular good or service.
Regression: A statistical process of relating one or more variables with another.
R-Squared: See Goodness of Fit
Two-Stage Least Squares (2SLS): A regression technique used when certain OLS assumptions
are violated.
U.S. DOT: The United States Department of Transportation
28 December 2007
Estimating Air Travel Demand Elasticities
Page i
Table of Contents
Executive Summary .........................................................................................................................i
Glossary of Terms/Abbreviations................................................................................................vii
1.0
Introduction ........................................................................................................................1
2.0
Elasticity Concepts ............................................................................................................2
3.0
Previous Evidence – Literature Review ...........................................................................8
4.0
Econometric Analysis......................................................................................................18
5.0
Conclusion: Synthesising the Findings.........................................................................32
1.1
2.1
2.2
2.3
2.4
3.1
3.2
4.1
4.2
4.3
4.4
4.5
4.6
5.1
5.2
5.3
5.4
Report Outline.........................................................................................................................1
Price Elasticity.........................................................................................................................2
Cross Price Elasticity. .............................................................................................................3
Income Elasticity.....................................................................................................................4
Demand Elasticities in the Context of Air Transportation........................................................4
Overview of the Literature.......................................................................................................8
Summary of the Papers Reviewed .........................................................................................9
Data Sources ........................................................................................................................18
Overview of Econometric Issues...........................................................................................19
Model Specifications .............................................................................................................21
Explanatory Variables ...........................................................................................................23
Econometric Results .............................................................................................................27
Econometric Results .............................................................................................................30
Introduction ...........................................................................................................................32
General Guidelines on Air Fare Elasticity Values .................................................................33
Comment on Income Elasticities...........................................................................................36
Conclusions ..........................................................................................................................38
Appendix A - Final Model Specifications ....................................................................................39
Appendix B - Omitted Variable Bias ............................................................................................41
Appendix C - Long-Run Elasticity in ARDL Model .....................................................................42
Appendix D - Regression Output.................................................................................................43
Appendix E - Data Sources ..........................................................................................................46
Appendix F - List of Countries by Region...................................................................................48
28 December 2007
Estimating Air Travel Demand Elasticities
Page 1
1.0 Introduction
This report documents the findings from extensive research undertaken by InterVISTAS Consulting
Inc. to determine air travel demand elasticities applicable to a wide range of air transport markets.
The research included:
ƒ
ƒ
ƒ
A literature review of previous research into air travel demand elasticities.
Comprehensive econometric analysis using three major datasets:
ƒ The U.S. DB1B database of domestic air travel 1994-2005.
ƒ The IATA PaxIS database of world air travel from 2005 onwards.
ƒ The UK International Passenger Survey (IPS) of outbound air travel from 2003 to 2006.
Development of a set of guidelines, and guideline elasticities, to be applied to the analysis of
different air markets, broken down by geographic market, length of haul and level of
aggregation (route, national and pan-national).
1.1
Report Outline
The report is structured as follows:
ƒ
ƒ
ƒ
ƒ
Chapter 2 provides a general discussion on elasticity concepts, particularly in the context of air
transportation.
Chapter 3 summarises the findings from a literature review of previous research work on air
transport demand elasticities.
Chapter 4 details the econometric analysis of three different dataset, including the elasticities
estimated.
Chapter 5 uses the results of the literature review and econometric analysis to develop
guidelines for the application of air fare elasticities.
Additional information on the study methodology and study findings is provided in the appendices.
28 December 2007
Estimating Air Travel Demand Elasticities
Page 2
2.0 Elasticity Concepts
In economics, elasticity measures the response or sensitivity of one economic variable to the
change in another economic variable. Elasticities are a useful concept as they allow decision
makers insight into the impact of different economic actions. A common elasticity concept is
demand elasticity. This measures the change in quantify demanded of a particular good or service
as result of changes to other economic variables, such as the price of the that good or service, the
price of competing or complimentary goods/services, income levels, taxes, etc.2
The section below outline three elasticity concepts that are relevant to this study: price elasticity,
cross price elasticity and income elasticity.
2.1
Price Elasticity
Price elasticity is a measure economists use to capture consumers’ sensitivity to price changes for
a particular good or service. The price elasticity is defined as:
Price Elasticity =
% Change in Quantity Demanded
% Change in Price
Since the quantity demanded generally decreases when the price increases, this ratio is usually
expected to be negative. However, sometimes analysts report the absolute value and therefore
the elasticity is often quoted as a positive number.3
As an example, suppose a good has a price elasticity of -0.6; a 10% increase in the price will result
in an 6% decline in the quantity demanded. For a good with a price elasticity of -1.2, a 10%
increase in the price will result in a 12% decline in the quantity demanded.
Goods with elasticities less than one in absolute value are commonly referred to as having inelastic
or price insensitive demand – the proportional change in quantity demanded will be less than the
proportional change in price. In this situation, increasing the price will increase the revenue of the
producer of the good, since the revenue lost by the relatively small decrease in quantity is less than
the revenue gained from the higher price.
Goods with elasticities greater than one in absolute value are referred to as having elastic or price
sensitive demand - the proportional change in quantity demanded will be greater than the
proportional change in price. A price increase will result in a revenue decrease to the producer
since the revenue lost from the resulting decrease in quantity sold is more than the revenue gained
from the price increase.
2
A equivalent concept is supply elasticity, which measures the change in the quantity supplied as a result of changes in other
economic variables, etc.
3
As the calculation uses proportionate changes, the result is a unitless number and does not depend on the units in which the price
and quantity are expressed. Therefore, elasticities for different goods or markets can be directly compared.
28 December 2007
Estimating Air Travel Demand Elasticities
Page 3
A number of factors affect the price elasticity of a good or service:
ƒ
ƒ
ƒ
ƒ
ƒ
Availability of substitutes: the more possible substitutes, the greater the elasticity. Note that the
number of substitutes depends on how broadly one defines the product. For example,
Chevrolet cars have a high price elasticity as they can be substituted by other brands of car
(Ford, BMW, Honda etc). If one considers the market for cars as a whole, the elasticity for cars
is lower as there are fewer substitutes (bus, taxi, cycling, etc).
Degree of necessity or luxury: luxury products tend to have greater elasticity. Some products
that initially have a low degree of necessity are habit forming and can become "necessities" to
some consumers. Bread has a low elasticity as it is considered a necessity, as does tobacco
because it is habit forming.
Proportion of the purchaser's budget consumed by the item: products that consume a large
portion of the purchaser's budget tend to have greater elasticity.
Time period considered: elasticity tends to be greater over the long run because consumers
have more time to adjust their behaviour. For example, short-term demand for gasoline is very
inelastic (approximately -0.2)4 as consumers have little choice but to continue consuming in
order that they can travel to work, school etc., although they can cut down on some leisure or
discretionary trips or use other modes. The long-term elasticity is higher (about -0.7 – still
inelastic) as consumers can purchase smaller cars, move nearer to work and other behavioural
changes in order to reduce consumption.
Whether the good or service is demanded as an input into a final product or whether it is the
final product (e.g., fuel is demanded as an input into production processes, transportation,
etc.). If the good or service is an input into a final product then the price elasticity for that good
or service will depend on the price elasticity of the final product, its cost share in the production
costs, and the availability of substitutes for that good or service.
2.2
Cross Price Elasticity.
The cross price measures the sensitivity of demand for a particular good to changes in the price of
another good:
Cross Price Elasticity =
% Change in Quantity of Good A Demanded
% Change in Price of Good B
When the elasticity is positive, the two goods are substitutes (e.g. Coca-Cola and Pepsi); when the
elasticity is negative the goods are compliments (e.g. coffee and milk).
4
Source: Economics: Private and Public Choice, James D. Gwartney and Richard L. Stroup, 1997
28 December 2007
Estimating Air Travel Demand Elasticities
2.3
Page 4
Income Elasticity
The income elasticity measures the sensitivity of demand for a good to changes in income:
Income Elasticity =
% Change in Quantity Demanded
% Change in Income
“Normal” goods have an income elasticity between 0 and +1: the quantity demanded increases at
the same or a lesser rate than the increase in income. For example, a good where a 10% increase
in income results in a 0-10% increase in consumption would be considered a “normal” good.
Goods with negative elasticities are referred to as inferior goods – as a person’s income increases
they buy less of that good and substitute with it with better quality goods (e.g. buy branded goods
rather than supermarket own-brands).
Luxury goods have an income elasticity above unity (one): consumption increases by a greater
proportion than income.5 Foreign travel is an example of such a good.
2.4
2.4.1
Demand Elasticities in the Context of Air Transportation
Level of Aggregation
In air transportation, as in many other sectors of the economy, the context in which the elasticities
are considered can affect the value of the elasticities. In particular, the elasticity can vary
depending on the availability of substitutes. In this study, five different levels of aggregation
(representing five different contexts) have been identified, as summarised in Figure 2-1.
Figure 2-1: Air Transport Elasticities and the Level of Aggregation
Level of Aggregation
Description
Fare Class Level
This the most disaggregate level. In this context, travellers are
choosing between different fare classes (first class, business class,
full economy, discount economy, etc.) on particularly airlines. At this
level, the elasticities are arguably highest. Travellers can easily
switch between fare-class levels, airlines, use of another mode of
travel (in some cases), or simply chose to not travel (i.e., other
activities act as a substitute for air travel). For example, in response
to an increase in the full economy fare on a given airline, the traveller
can respond by booking a discount economy fare on the same airline,
or book with another airline, or travel by another mode.
Carrier Level
The elasticities at this level of aggregation reflect the overall demand
curve facing each air carrier on a given route. In situations where
there are a number of air carriers serving the route, the demand
5
“Luxury good” is a term of art in economics that merely indicates that demand increases more than proportionately with income.
Education, medical services, entertainment and other goods fit this category. The vernacular use of the term “luxury” may have
value judgements attached, which are not intentional in the context of this report.
28 December 2007
Estimating Air Travel Demand Elasticities
Page 5
Level of Aggregation
Description
elasticity faced by each carrier is likely to be fairly high - if an air
carrier increases it fare unilaterally, it is likely to lose passengers to
other carriers operating on that route.6
Route/Market Level
At the route or market level (e.g., Heathrow–Paris CDG or LondonParis), the elasticity response might be expected to be generally
lower than at the fare class or carrier level. Travellers faced with a
fare increase on all carriers serving a route (e.g., due to an increase
in airport fees and charges), have fewer options for substitution.
However, they can chose to travel on an alternative route (e.g., travel
Gatwick-CDG or Heather-Orly rather than Heathrow-CDG), travel by
another mode (in some cases), or not travel.
National Level
At the national level, fare elasticities would be expected to be lower
still, as travellers have fewer options for avoiding the fare increase.
For example, if a national government imposed a new or increased
tax on aviation, travellers could only avoid this increase by using
another mode (which may not always be possible), or not travelling
(or possibly travelling elsewhere). For example, if the UK government
imposed an increased tax on aviation, UK residents travelling to
mainland Europe could respond by travelling by Eurostar or ferry, or
by not travelling. Similarly, travellers in France could respond to the
tax by travelling to the UK by rail/ferry or by travelling to another
country, such as Germany or Spain.
Pan-National Level
This represents the most aggregate level considered, in which a fare
increase is imposed at some pan-national level; for example, the
European Union imposing an aviation tax on all its member states. In
this case, the options for avoiding the fare increase are even further
reduced, so therefore the elasticity would be expected to be lower.
Given the economic and policy focus of this study, the first two levels of aggregation (fare class
and carrier level) are outside the scope of this study. This study focuses largely on elasticities
appropriate for the route, national and pan-national level of aggregation.
2.4.2
Cross Price Elasticites
In noted previously, cross price elasticity measures the response in demand for a given good or
service to changes in the price of substitute or complimentary goods and services. In each of the
five levels of aggregation discussed in the previous section, differing cross elasticities exist,
reflecting the various substitute options available. For example:
6
Even in a situation where air carrier has a monopoly on a route, it may still face a fairly high demand elasticity as connecting
options can also act as a substitute. E.g., travel London-Delhi via Paris or Dubai rather than direct.
28 December 2007
Estimating Air Travel Demand Elasticities
ƒ
ƒ
ƒ
ƒ
ƒ
ƒ
ƒ
Page 6
At the fare class level, an increase in the price of the full economy fare could increase the
demand for both business class tickets and discount tickets.
At the carrier level, an unilateral increase in the fare of one particular carrier on a route can
increase the demand for other carriers on the route (and the demand for connecting
alternatives).
At the route level, an increase in the cost of travel from Heathrow to CDG can increase the
demand for travel on Gatwick-CDG or Heathrow-Orly.
At the national level, an increase in the cost of air travel to/from a given country may increase
demand for air travel to/from other substitute countries.
At the pan-national level, an increase in the cost air travel to/from a particular region may
increase demand for air travel to/from other regions. E.g., an increase in the cost of air travel to
the EU may increase demand for air travel to the U.S.
At all levels of aggregation, there may exist cross elasticity effects with other modes of
transport. An increase in the cost of air travel may increase demand for ground transportation
and vice versa.
There may also cross elasticity effects between air travel and other leisure or consumption
activities.
The degree of cross elasticity effects will differ depending on the specific circumstances involved,
and in some cases may not exist at all (e.g., in long-haul markets there generally are no substitutes
for air travel). It is also worth noting that while the examples above focus on substitution effects
(i.e., positive cross elasticities), there may also exist complimentary (i.e., negative) cross
elasticities. For example, the price of hotels may impact on the demand for air travel.
It can also be shown that the own price elasticity at one level of aggregation reflects both the own
price and cross price elasticities at other levels of aggregation. For example, the price elasticity at
the route level is a function of the own price and cross price elasticities at the fare class and carrier
levels of aggregation.
Clearly, the cross elasticity effects and the interaction between own price and cross price
elasticities at different levels of aggregation adds significant complexity to any analysis of air travel
demand elasticities. Any econometric analysis of air fare elasticities needs to be clear as to which
own price and cross price elasticity were being measured and controlled for. An analysis of price
elasticities which does not include cross elasticity terms may represent a price elasticity at a
different level of elasticity. For example, an analysis of route-level elasticities which does not
control for route substitution effects may be more akin to national-level elasticity.
2.4.3
Air Fares and Total Travel Costs
Some researchers have questioned whether estimated air fare elasticities are consistent with the
total cost of travel (which may also include hotel, ground transport to/from the airport, food,
28 December 2007
Estimating Air Travel Demand Elasticities
Page 7
entertainment, etc.) and whether any analysis of air fare elasticities should also include total travel
costs.7 The argument around this issue can be characterised as follows:
Air transport typically represents approximately 25% of the total travel costs associated
with leisure travel (the exact percentage varies depending on the length and type of travel).
This percentage has declined in recent years, due to the development of Low Cost
Carriers (LCCs). If, say, the air fare elasticity is around unity then a 10% increase in the
air fare will reduce demand for travel by 10% (if the demand for air travel declines by 10%,
it is anticipated that the demand for the other components of the travel package will also
decline by 10%).
However, a 10% increase in the air fare represents a 2.5% increase in the total cost of
travel. This then implies that the demand elasticity with respect to total travel costs is:
-10% / 2.5% = -4.0
This represents an extremely high elasticity and one which is not matched by quantitative
research of tourism demand elasticities (which generally find much lower elasticities).
Therefore, there appears to be an inconsistency between the size of price elasticities
estimated for the air transport industry and those estimated for the overall travel and
tourism industry.
However, there two explanations for this apparent inconsistency:
ƒ
ƒ
The approach above to calculating the tourism elasticity is applicable only where there are
limited opportunities for substitution. In situations where one component of a package can be
substituted for an adequate alternative then the price elasticity for that component can be
much higher than suggested by the price elasticity of the overall package. For example, the air
transport component could be substituted with air travel on another route (e.g., flying GatwickCDG rather than Heathrow-CDG), travel on another mode, or air travel to another destination
(e.g., flying to Italy for a ski vacation rather than flying to France).
There is some evidence that some travellers do not fully consider the total cost of their travel
but instead apply a “two-stage” process in their decision making. Travellers are induced to
select a destination based on the level of air fare offered, and having booked the flight then
consider the other costs associated with the travel. For example, LCCs such as Ryanair offer
seat-sales with zero or very low costs (e.g., 1 pence) which have been successful in attracting
passengers despite the fact that passengers will incur substantial additional costs associated
with airport parking, ground transportation, accommodation, etc. While not all travellers
engage in this “two-stage” process, a significant proportion may do so, which is reflected in the
air fare elasticities.
7
For example, see the UK Civil Aviation Authority report, Demand for Outbound Leisure Air Travel and its Key Drivers, December
2005.
28 December 2007
Estimating Air Travel Demand Elasticities
Page 8
3.0 Previous Evidence – Literature Review
3.1
Overview of the Literature
A literature of previous research into air fare demand elasticities was undertaken in order to
provide a greater understanding of air fare elasticities and to aid the econometric analysis
conducted later in the study. A review of this nature enables the econometrician to obtain insight
into how different aspects of markets may affect results; it also can provide conceivable solutions
to standard and non-standard issues that arise. In total, 22 studies were reviewed, including twometa analyses of multiple publications. Most papers were retrieved from peer reviewed academic
publications or were the product of government-related commissions. Summaries of these studies
are provided in Section 3.2.
As can be seen in the table, the research produced a wide range of air fare elasticity estimates,
depending on the markets analysed, time period, methodology and data. This illustrated by the
meta-study conducted in 2002 by Gillen, Morrison, Stewart, Air Travel Demand Elasticities:
Concepts Issues and Measurement. The range of elasticities found by the authors is summarised
in Figure 3-1. The figure shows the range of values estimated in the studies surveyed and the
most-likely value (the black dot) determined by Gillen et al. This meta-study found elasticities
ranging from -0.198 to -1.743, depending on the market.
Figure 3-1: Own-Price Elasticities of Demand (Gillen et al., 2002)
More Elastic
-2
-1.5
Less Elastic
-1
Long-haul international business:
-0.5
0
-0.475
-0.198
-0.265
Long-haul international leisure:
-1.7
-0.56
-1.04
Long-haul domestic business:
-1.428
-0.836
-1.15
Long-haul domestic leisure:
-1.228
-0.787
-1.104
Short-haul business:
-0.783
-0.595
-0.7
Short-haul leisure:
-1.743
-1.288
-1.520
Reproduced from Air Travel Demand Elasticities: Concepts, Issues and Measurement, D. Gillen, W.G. Morrison and C. Stewart,
2002.
28 December 2007
Estimating Air Travel Demand Elasticities
Page 9
Although there is a considerable range of elasticities, some generalised findings can be drawn from
the literature review:
Sensitivity to Air Fare Changes
All of the studies reviewed, spanning a period of over 25 years, found that there was a significant
demand response to changes in air fares, such that increases in air fare lead to lower passenger
traffic demand. The consistency of this result strongly indicates that any policy action that results
in higher fares (e.g., taxes, increased landing fees) will result in a decline in demand. The actual
decline in demand will depend on a number of factors, as discussed below.
Business Versus Leisure Passengers
In general, the results show that, all else being equal, business travellers are less sensitive to fare
changes (less elastic) than leisure travellers. Intuitively, this result is plausible – business travellers
generally have less flexibility to postpone or cancel their travel than leisure travellers.
Nevertheless, the studies do show that even business travel will decline in the face of fare
increases, albeit not to the same extent as leisure travel.
Short-Haul Versus Long-Haul Travel
Another fairly consistent finding was that fare elasticities on short-haul routes were generally higher
than on long-haul routes. In part, this reflects the opportunity for inter-modal substitution on short
haul routes (e.g., travellers can switch to rail or car in response to air fare increases).
Carrier Versus Market Versus National Elasticities
Some of the studies supported the concept that the demand elasticity faced by individual air
carriers is higher than that faced by the whole market. For example, Oum, Zhang, and Zhang
(1993) estimated firm-specific elasticities in the U.S. and estimated values ranging from -1.24 to
-2.34, while studies estimating market or route elasticities ranged from -0.6 to -1.6. In contrast,
Alperovich and Machnes (1994) and Njegovan (2006) used national-level measures of air travel in
Isreal and the UK respectively and produced even lower elasticity values (-0.27 and -0.7,
respectively).
Income Elasticities
Many of the studies also including an income variable, in part, to isolate the effects of a shift along
the demand curve (such as would be caused by a price change) from the effect of a shift in the
demand curve (as might be caused by a change in income levels). The studies including the
income term all produced positive income elasticities, as would be expected (air travel increases as
incomes increase). Virtually all of these studies estimated income elasticities above one, generally
between +1 and +2. This indicates air travel increases at a higher rate than incomes.
3.2
Summary of the Papers Reviewed
The table in Figure 3-2 provides a summary of each of the papers reviewed, including the elasticity
estimated produced, dependent and explanatory variables used, and a short description of the
methodology and key findings.
28 December 2007
Estimating Air Travel Demand Elasticities
Page 10
Figure 3-2: Summary of the Air Fare Elasticity Literature Review
Author / Paper
Elasticity
Estimates
Dependent
Variables
Explanatory
Variables
Findings
Taplin - A Coherence Approach to
Estimates of Price Elasticities in
the Vacation Travel Market (1980)
Price (leisure):
-0.9 to -3.3
Results
synthesized from
other studies.
Results synthesized from
other studies.
Cross-elasticities of various substitute and compliment
goods were inferred based on the results observed in
other studies. Accommodations, domestic travel, car
costs, and prices of other consumer goods were
analysed as for effects on foreign air travel.
Abrahams - A Service Quality
Model of Air Travel Demand: An
Empirical Study (1983)
Price:
-0.36 to -1.81
Expected
schedule delay
time. Price
elasticity of
demand is
calculated
indirectly.
Traffic.
lowest unrestricted fare.
product of city pair
populations.
% change in GNP.
perceived price of air
transport relative to auto
transport.
Found rapid growth of hotel and recreation facilities in
Hawaii in response to the introduction of low cost jet
service; expansion of business activity in Reno as a
result, in part, of the sharply increased service quality in
airline services.
Income:
1.0 to 2.6
Income:
0.46 to 1.6
Long-haul routes appeared to be more elastic than short
and vacation traffic to be more elastic than business
traffic.
A negative correlation was found between the reduction
in 1980 fare levels from the official CAB fare and the
estimated service quality elasticity.
Oum, Gillen, and Noble - Demands
for Fareclasses and Pricing in
Airline Markets (1986)
28 December 2007
Price: -1.152
(all routes)
Income: -1.445
(all routes)
Route aggregate
demand.
Average fare.
Per capita incomes
between city pairs.
Population between city
pairs.
Vacation route dummies.
Derived partial elasticities for three fare classes using a
translog demand system in a first stage then the second
stage involves estimating a log-lin demand function to
measure total price elasticities. Ramsey-optimal fare
class prices were also computed by minimizing
estimated airfare index functions subject to breakeven
constraints. Intra-U.S. routes were used as data
sources.
Estimating Air Travel Demand Elasticities
Page 11
Author / Paper
Elasticity
Estimates
Dependent
Variables
Explanatory
Variables
Findings
Oum - Alternative Demand Models
and their Elasticity Estimates
(1989)
No aviation
elasticities
estimated.
No aviation
elasticities
estimated.
N/A
Examined linear demand, Log-Lin demand, Box-Cox,
logit, and translog demand models for effectiveness in
demand analysis and forecasting. The translog model
was shown to be the model that produces the most
reasonable results – elasticities exhibit stability and
predictability, smaller standard errors than from log-lin
model.
Oum, Zhang, and Zhang - InterFirm Rivalry and Firm-Specific
Price Elasticities in Deregulated
Airline Markets (1993)
Price:
-1.24 to -2.34
on domestic U.S.
routes
Route aggregate
demand by
carrier.
Average fare.
Total income.
Seasonality dummies.
Vacation route dummies.
Cost per passenger mile.
Firm specific price elasticities were measured, and were
found to increase with distance. Vacation routes were
found to have higher elasticity values. In addition, the
analysis found that firms were shown to behave uniquely
in a duopoly environment.
Alperovich and Machnes - The
Role of Wealth in the Demand for
International Air Travel (1994)
Price: -0.27
(all routes)
Travelers per
capita.
Financial assets.
Non-financial assets.
Wages.
Consumer price index.
Authors examine air travel out of Israel. Price was found
to be inelastic while income was highly elastic. Used
log-lin models. Inclusion of wealth variables is found to
reduce serial correlation, correct bias, and improve
estimate precision. Total assets (including financial and
non-financial assets) were determined to be significant in
demand.
Australian Bureau of Transport and
Communications Economics Demand Elasticities for Air Travel
to and from Australia (1995)
Price:
-0.14 to -1.19
(Aus. leisure)
Total O/D leisure
passenger; Total
O/D business
passengers
Real household
disposable income.
Price index of domestic
holiday travel and
accommodations.
Annual average
exchange rates.
Relative prices of holiday
travel and
accommodations.
Airfares, income and relative prices found to be
important determinants of leisure travel to and from
Australia. Income and relative prices were important for
business travel. Real exchange rate elasticities are also
examined. Airfare elasticities differed between
passenger type and O/D market. Linear and Log-Log
models were employed.
Income:
1.64 to 2.06
-0.5 to -1.86
(foreign leisure)
-0.01 to -0.4
(Aus. Business)
28 December 2007
Estimating Air Travel Demand Elasticities
Author / Paper
Elasticity
Estimates
Page 12
Dependent
Variables
Explanatory
Variables
Findings
Per capita figures are
used to account for
population effects.
-0.16 to -0.62
(Foreign
business)
Income:
0.21 to 11.58
(Aus. leisure)
1.88 to 5.51
(foreign leisure)
Cohas, Belobaba, and Simpson Competitive Fare and Frequency
Effects in Airport Market Share
Modeling (1995)
Price:
-0.37 to -0.83
Airline market
share
Quarterly
Origin/Destination traffic
data from 1979/1 to
1992/2.
Paper looked at competitive effects in duopoly markets.
The elasticity of market share with respect to frequency
of service was positive; the direct elasticity of market
share with respect to price was negative; the crosselasticity of market share with respect to price was
positive.
Jorge-Calderon - A Demand Model
for Scheduled Airline Services on
International European Routes
(1997)
Price:
-0.534
(all economy)
Total scheduled
traffic between
two cities. Taken
from ICAO Traffic
by Flight-Stage
Survey.
Population of O/D cities.
Incomes of O/D cities.
Distance.
Frequency.
Average aircraft size.
Unrestricted economy
fares.
Dummies: Travel over
sea water; proximity of
nearby hub airport;
discounted restricted
fares; holiday resort
destination.
International European routes were examined using a
demand model that used several independent variables.
Various stage lengths were examined separately.
Overall, demand was shown to be price inelastic with a
tendency for elasticities to increase with distance but fall
in the long-haul sector. Highly discounted fares have a
positive effect on traffic. Discounted fares were used
more often in short-haul markets (presumably to
compete with other modes); longer distance flights were
more price sensitive due to the reduced use of
discounted fares.
Frequency:
0.79 to 1.26
Aircraft Size:
0.55 to 1.74
28 December 2007
Estimating Air Travel Demand Elasticities
Page 13
Author / Paper
Elasticity
Estimates
Dependent
Variables
Explanatory
Variables
Findings
Taplin - A Generalised
Decomposition of Travel-Related
Demand Elasticities into Choice
and Generation Components
(1997)
Price:
-1.7 to -2.1
(leisure)
Estimates from
Taplin (1980)
Estimates from Taplin
(1980)
Firm specific price elasticities were measured, and were
found to increase with distance. Vacation routes were
found to have higher elasticity values. Cross elasticities
between domestic and international vacation choices
were examined. Expenditure choice and generation
elasticities were derived separately.
Hamal - Australian Outbound
Holiday Travel Demand Long-haul
Versus Short-haul (1998)
Price:
-0.35 to -2.23
Short-term
resident
departures for
holiday purposes.
Real per capita
household income.
Price index of domestic
holiday travel and
accommodation over
domestic CPI.
Price index of foreign
country holiday travel
and accommodation over
foreign CPI.
Foreign/domestic
exchange rate.
Exchange rate weighted
by prices of overseas
and domestic travel and
accommodations.
Paper makes use of four log-log models with different
explanatory variable combinations to measure
elasticities for travel demand to various markets outside
of Australia. Income elasticities were shown to vary
depending on the market. Cross price elasticity with
domestic demand and accommodations were positive
and above one for all markets.
Carlsson - Private vs. Business
and Rail vs. Air Passengers:
Willingness to pay for Transport
Attributes (1999)
Price:
-1.09 to -1.43
(total)
Number of trips.
Elasticities were inferred
indirectly through the use
of a logit model that
accounts for mode
choice decision making.
A stated preference survey was used to generate data
for passenger’s willingness to pay for improvements to
various transport modes through a conditional logit
model. Routing were limited to travel between
Stockholm and Gothenburg, Sweden. Air Arlanda and
Air Bromma estimates were generated separately.
Business travelers are found to value time more highly
and were less price elastic than private passengers.
Income:
1.1 to 2.1
Income:
0.63 to 0.84
-0.94 to -1.28
(business)
-2.95 to -3.04
(personal).
28 December 2007
Estimating Air Travel Demand Elasticities
Page 14
Author / Paper
Elasticity
Estimates
Dependent
Variables
Explanatory
Variables
Findings
Abed, Ba-Fail, Jasimuddin - An
Econometric Analysis of
International Air Travel Demand in
Saudi Arabia (2001)
No elasticity
estimates.
Demand for
international air
travel.
Population size.
Expenditures.
A proposed econometric model of demand was derived
for Saudi Arabia international air travel. Population and
expenditures were found to be the primary determinants
of international air travel in Saudi Arabia.
Gillen, Morrison, Stewart - Air
Travel Demand Elasticities:
Concepts Issues and Measurement
(2002)
Price: -0.27
(long-haul int.
business)
Survey of a large
group of studies.
Survey of a large group
of studies.
The report was based on an extensive survey of
literature related to provide air travel elasticity estimates.
Six distinct markets for air travel were identified:
business and leisure travel; long-haul and short-haul
travel; and international and North American long-haul
travel. Estimates vary significantly, reflecting the range
of studies that were examined.
-1.04
(long-haul int.
leisure)
-1.15
(long-haul dom.
business)
-1.10
(long-haul dom.
leisure)
-0.7
(short-haul
business)
-1.52
(short-haul
leisure)
Income:
1.39
Median Values
Reported
28 December 2007
Estimating Air Travel Demand Elasticities
Page 15
Author / Paper
Elasticity
Estimates
Dependent
Variables
Explanatory
Variables
Findings
Brons, Pels, Nijkamp, Rietveld Price Elasticities of Demand for
Passenger Air Travel: A MetaAnalysis (2002)
No direct
measures.
Price-elasticity.
Transfer distance.
Fare Class.
Geographic location.
Research method (time,
cross-section, or pooled).
Time horizon.
Period of data collection
This paper is a meta-analysis of the factors affecting
price elasticities in the aviation sector. Long-run price
elasticities were higher in absolute value; passengers
became more price sensitive over time; Business
passengers were less sensitive to price – the difference
is about 0.6; European passengers were not more price
sensitive than U.S. passengers and Australian
passengers.
New Zealand Commerce
Commission - Final Report Part IV
Inquiry into Airfield Activities at
Auckland, Wellington and
Christchurch International Airports
(2002)
Price:
-1.3 (domestic)
Estimates
obtained from
other studies.
Estimates obtained from
other studies.
Report cites that the price elasticity of the derived
demand by airlines for airfield services can be inferred
from the elasticity of the demand for airline travel –
requires an assumption made about what portion of any
change in landing charges is passed to passengers by
airlines.
Castelli, Pesenti, Ukovich - An
Airline-Based Multilevel Analysis of
Airfare Elasticity for Passenger
Demand (2003)
Price:
-1.058
(all routes)
Number of
passengers
travelling on a
route, in fare
class, on a given
day. No
distinction is made
between origin
and destination.
Fare.
Population of the total
metropolitan are served
by airports.
GDP per capita in the
two airport catchment
areas.
Distance between the
two airports.
A measure of the cost
faced by travellers in
other modes of
transportation.
Daily frequency of flights.
Aircraft size.
Hub (dummy).
Price elasticity of a specific airline (Air Dolimiti – the
largest Italian regional carrier) was estimated. Nine
routes were examined, price elasticity was found to vary
significantly across the various routes – from -0.75 to
-1.62.
-1.8 (international)
Ranged from
-0.75 to -1.62 on
specific routes
Frequency:
0.862
28 December 2007
Estimating Air Travel Demand Elasticities
Page 16
Author / Paper
Elasticity
Estimates
Dependent
Variables
Explanatory
Variables
Findings
PriceWaterhouseCoopers Aviation Emissions and Policy
Instruments, Final Report (2005)
Price:
-0.73
(business)
No information
provided.
No information provided.
Impacts on competitiveness and economic performance
were estimated for the European Union based on the
prospected introduction of certain environmental policy
changes. The authors conducted their own estimates of
elasticities but reject their estimates in favour of figures
derived from Gillen et al. (2001).
No information
provided.
No information provided.
Authors postulate that demand for air travel has become
more elastic with the advent of online purchasing making
prices more transparent – heightened competition and
increased awareness.
-1.52
(leisure)
-1.23
(full service)
-1.38
(low cost)
-1.02
(cargo)
Estimates derived
from Gillen et al.
(2001)
Rubin and Joy - Where are the
Airlines Headed? Implications of
Airline Industry Structure and
Change for Consumers (2005)
28 December 2007
Price:
-2.4
(leisure)
Estimates from
1997 study –
Mackinac Center
for Public Policy,
Price Elasticity of
Demand
Due to the high price elasticity for leisure travel, airlines
pass these charges forward as surcharges to
consumers.
Estimating Air Travel Demand Elasticities
Page 17
Author / Paper
Elasticity
Estimates
Dependent
Variables
Explanatory
Variables
Findings
Goolsbee and Syverson - How Do
Incumbents Respond to the Threat
of Entry? Evidence from the Major
Airlines (2006)
Price:
-0.64 to -1.12
Total passengers
or mean fares.
DB1A files from
Q1 1993 to Q4
2004.
Time dummies:
Southwest establishing
presence at both
endpoints of route w/o
flying route; Southwest
flying route.
The paper shows that the threat of Southwest entering a
market was sufficient to encouraging incumbents to
lower their prices – this was also said to cause an
increase in demand prior to Southwest beginning
service. The fare and quantity changes from this period
implies a demand elasticity between -0.64 and -1.12.
Various control variables
are used in the different
model specifications.
Njegovan - Elasticities of Demand
for Leisure Air Travel: A System
Modelling Approach (2006)
28 December 2007
Price: -0.7
(all routes)
Income: 1.5
(all routes)
Share of
household budget
spent on leisure
air travel.
Price of air travel.
Price of tourism abroad.
Price of domestic
tourism.
Total expenditures on
leisure.
Analysis of leisure travel demand elasticities in the
United Kingdom. Estimated that domestic leisure market
has income elasticity of 0.6. Elasticity with respect to air
fare changes is inelastic. The cross-price elasticities in
the air travel equation were relatively large compared to
the value of the own-price elasticity. The finding of a
relatively low aggregated market own-price elasticity is
not inconsistent with some relatively large own-price
elasticities which are estimated from route-specific data
where low cost airlines have been successful in
attracting large volumes of traffic by offering low fares.
Estimating Air Travel Demand Elasticities
Page 18
4.0 Econometric Analysis
4.1
4.1.1
Data Sources
U.S. Domestic (DB1B)
U.S. domestic market data was obtained from the U.S. DOT’s Data Base 1B (DB1B). The DB1B
data is widely used throughout the aviation industry and as a result any errors are generally noticed
very quickly and brought to the attention of the U.S. DOT for correction. Also, the database has
been used extensively by other researchers to generate reasonable estimates of demand
elasticities. As a result, the DB1B data is a very reliable source of data.
Within the U.S. all large domestic carriers are required to provide the DOT with information about
every tenth itinerary, for both domestic and international flights, to the U.S. DOT.8 Thus, the DB1B
represents a 10% sample of origin-destination passengers and airfare for each airline on each
route. Quarterly traffic and fare data was taken for the period from 1994 Q1 to 2005 Q4 for the top
1000 city pair routes (by traffic). Traffic figures reflect the actual number of passengers on a
particular route during the given quarter.9 Average fare reflects the estimated (based on a 10%
sample) average one-way fare paid (in USD) on a particular route.
4.1.2
World Regions (PaxIS)
The analysis of world regions was done using IATA’s Passenger Intelligence Service (PaxIS)
database. The database captures market data through IATA’s Billing and Settlement Plan (BSP)
and uses statistical estimates to address missing direct sales, low cost carriers, charter flight
operators, under-represented BSP markets, and non-BSP markets. The advantage of data series
is that traffic and fare estimates are available for routes around the world. However, this dataset is
limited by the short time series (data is available from 2005) and the comparative lack of use in
prior studies.
For the purposes of this report, select regions were chosen to describe the distinct marketplaces
that exist in modern air travel:10
ƒ
Intra-Europe;
ƒ
Intra South Asia and South East Asia;
ƒ
Trans Atlantic (Europe to/from North America);
ƒ
Trans Pacific (South Asia and South East Asia to/from North America) ;
8
A large carrier is defined as an air carrier holding a certificate issued under 49 U.S.C.41102, as amended, that: (1) Operates
aircraft designed to have a maximum passenger capacity of more than 60 seats or a maximum payload capacity of more than
18,000 pounds; or (2) conducts operations where one or both terminals of a flight stage are outside the 50 states of the United
States, the District of Columbia, the Commonwealth of Puerto Rico and the U.S. Virgin Islands.
9
Since data comes from a 10% sample the actual number of passengers can be accurately calculated simply by multiplying the
number of passengers reported in the DB1B by a factor of 10.
10
Further description of the countries included in the various regions can be found in Appendix F.
28 December 2007
Estimating Air Travel Demand Elasticities
Page 19
ƒ
Intra Sub Sahara Africa (Central/Western Africa and Eastern Africa);
ƒ
Intra Latin America (Non-Caribbean Central America and South America);
4.1.3
UK Outbound (IPS)
The data set used for analysis of the UK outbound travel market was the UK International
Passenger Survey (IPS). The International Passenger Survey is a face to face survey of a random
sample of passengers entering or leaving the UK by air, sea or the Channel Tunnel. In terms of air
passenger surveying, regular staggered shifts are scheduled at Heathrow’s four terminals,
Stansted’s two terminals, and Manchester’s two terminals. In addition, a number of small shifts are
scheduled on a quarterly basis at the other international UK airports. The sampling technique
represents about 1 in every 500 passengers and reflects about 90% of all travelers into and out of
the UK.
This report exclusively used outbound to Western Europe leisure air passenger traffic data from the
IPS. Quarterly traffic and fare data was taken for the period from 2003 Q2 to 2006 Q2. Traffic
figures reflect the estimated number of passengers on a particular route during the given quarter.
Average fare reflects the estimated average fare paid (in GBP) on a particular route, in the given
quarter.
4.2
4.2.1
Overview of Econometric Issues
Between, within and total variation
The data in each database consists of observations on city pair passenger itineraries for each
quarter (or month, in the case of PaxIS). These time-series cross-section databases are referred
to as panel data sets.
Panel data has two primary types of variation. Between variation is that based on differences
between the cross sectional units (OD pairs). Within variation is based on variation over time for a
given cross sectional unit.
It is important to recognise the different variances and their implication for estimation. For
example, a classic study of airline costs (data is by airline for a number of time periods) reveals
data with the following pattern:
unit
cost
traffic (quantity)
28 December 2007
Estimating Air Travel Demand Elasticities
Page 20
If a regression line is estimated with using all the data (total variation), then it may look like the
following:
unit
cost
This regression would find constant returns to scale.
traffic (quantity)
An alternative to estimation with total variation is to use within variation. This can be done, for
example, by using indicator (dummy) variables for air carriers or by doing the regression only using
data for a single route.11 If the regression uses within variation via dummy variables, then
estimated relationship will look like diagram below, which reveals the presence of economies of
traffic density. That is, the dummy variable for the airline controls for differences between the
carriers (such as size of network) and reveals how unit costs change with traffic within a given
network.
unit
cost
traffic (q uantity)
11
Another approach may be to introduce a lagged value of the dependent variable into the regression.
28 December 2007
Estimating Air Travel Demand Elasticities
Page 21
Estimation can be done using between variation, e.g., by introducing a measure of network size, or
by doing the regression on data for a single time period but with data for different airlines. In this
case, the regression results might look like the following:
unit
cost
traffic (quantity)
This regression might suggest the presence of mild diseconomies of network size. The main point
is that revealing the true cost relationship with panel data can be challenging, and simple
regressions may be misleading of the true phenomena.
4.3
Model Specifications
To estimate the elasticities of air travel, econometric analysis of the data was undertaken. A
number of different model specifications were tested to provide explanations of economic
phenomena within each given market. In all models, traffic was the dependent variable. In the
DB1B data set, traffic was taken at the city pair level (e.g., New York-Los Angeles). In the IPS and
PaxIS data sets, traffic was taken at the airport pair level (e.g., JFK-LHR, CDG-FRA). The subsections below describe the broad set of models employed in this analysis.
4.3.1
Ordinary Least Squares
Ordinary Least Squares (OLS) regression analysis is a method relating passenger traffic to air
fares, income (GDP) levels, and other variables that have an intuitive and measured impact on air
travel, while minimizing the variance (randomness) of the estimates. The regression analysis
allows the relationship between traffic and airfares to be isolated and quantified while controlling for
other factors that may impact air travel, such as gross domestic product, population levels, route
distance, seasonality, etc.
28 December 2007
Estimating Air Travel Demand Elasticities
Page 22
The OLS models used a log-log formulation, as follows:12
ln(Traffic) = Constant + a1 x ln(Fare) + a2 x ln(Var2) + … + an x ln(VarN) + an+1 x Dummies
Where:
ƒ
ƒ
ƒ
ƒ
ƒ
Traffic is the dependent variable,
Fare is the average economy or leisure fare.
Var2 to VarN are other quantifiable explanatory variables that affect traffic levels.
ln() refers to the variables inside of the parentheses transformed by the natural logarithm.
The dummies are variables that take the form of 1 or 0 in any observation and capture any
remaining structural reasons for traffic differences between routes. For example, a dummy for
the month of July would take the value of 1 for any observations from July and 0 during all
other months’ observations.
The regression analysis estimates the value of the parameters (constant, a1, a2, a3, a4, a5, etc.) on
each of the variables, which reflect the relative impact of each of the variables on traffic levels. As
log formulations approximate percentage changes in impacts, the parameters of the logged
independent variables can be directly interpreted as elasticities.13
4.3.2
Two-Stage Least Squares
Two-stage least squares (2SLS) is a regression technique that is used when explanatory variables
are believed to be correlated with the regression model’s error term used to obtain consistent
estimators.14 One or more instrumental variables (IVs) that are correlated with the endogenous
explanatory variable, but uncorrelated with the dependent variable are used to isolate the effects of
the endogenous explanatory variable. This process increases consistency (relative to OLS), at the
expense of increasing sample variance.
In this study, InterVISTAS experimented with the use of two-stage least squares techniques to
improve the consistency of elasticity estimates. The natural logarithm of distance (distance) and
the natural logarithm of fuel prices (fuel prices) were used separately and combined as potential
IVs. In some data sets, distance was found to be a worthwhile IV, exhibiting high correlations with
fares and low correlations with traffic. However, there is some concern that distance should be
used as an explanatory variable instead of an instrument (if route distance is believe to have an
impact on traffic). The reasonable use of distance as an IV is described in section 4.4. Fuel prices
were found to be poor IVs. Fuel prices exhibited low correlations with traffic and fares.
4.3.3
Autoregressive Distributed Lag
An autoregressive distributed lag (ARDL) model was also developed and used with OLS
regression analysis. Traffic was regressed onto similar explanatory variables as in prior models,
12
Log-log model formulations refer to a model specification where both the dependent (left hand side) and independent (right hand
side) variables have been transformed by the natural logarithm. Additional regression analysis results are provided in Appendix D.
13
This formulation is also referred to as a Cobb-Douglas functional form.
14
A consistent estimator in regression analysis is defined as an estimator that converges in probability to the correct parameter as
the sample size grows.
28 December 2007
Estimating Air Travel Demand Elasticities
Page 23
but also onto lagged values of traffic. The inclusion of lagged values of the dependent variable
(traffic) is done to account for stickiness, that is, the slow adjustment of supply (in the form of
capacity) to changes in the explanatory variables. The applicability of this assumption is less
reasonable for the U.S. domestic market as the barriers to expanding capacity are fewer than on
international routes; the U.S. domestic market has therefore been excluded from ARDL
estimation.15
InterVISTAS experimented with the use of ‘prior period’ and ‘year over year’ lags. Although both
showed some degree of success in controlling for these factors, ‘year over year’ lags had higher
correlations with current traffic levels than ‘prior period’ lags, and were determined to be the
preferred form of lagged variable.
The ARDL models used a formulation as follows:16
ln(Traffic t) = Constant + b1 x ln(Traffic t-1) + b2 x ln(Fare t) + b3 x ln(Var3 t) + … + bn x ln(VarN t)
+ bn+1 x Dummies t
Where:
Traffic t is the dependent variable,
ƒ Traffic t-1 is the traffic in the same month (or quarter) of the previous year
ƒ Fare is the average economy or leisure fare.
ƒ Var2 to VarN are other quantifiable explanatory variables that affect traffic levels.
ƒ ln() refers to the variables inside of the parentheses transformed by the natural logarithm.
ƒ The dummies are variables that take the form of 1 or 0 in any observation and capture any
remaining structural reasons for traffic differences between routes. For example, a dummy for
the month of July would take the value of 1 for any observations from July and 0 during all
other months’ observations.
Since traffic appears on both sides of the equation, the coefficients on the explanatory variables
cannot be directly interpreted as long-run elasticities (as is the case in the other OLS and 2SLS
models). The long-run elasticities are defined as when traffic across time periods stabilize.17 In
general, the use of ARDL models tended to produce more elastic fare elasticity estimates with
much higher goodness of fit values.
ƒ
4.4
Explanatory Variables
The following section describes the explanatory variables used in the regression analysis. Data
sources are provided in Appendix F.
15
The U.S. domestic market is not constrained by air bilateral agreements, has a wide access to capital, and high levels of
competition, all of which increase responsiveness to changes in the explanatory variables (fare price, income, and population).
16
Log-log model formulations refer to a model specification where both the dependent (left hand side) and independent (right hand
side) variables have been transformed by the natural logarithm. Additional regression analysis results are provided in Appendix C.
17
The derivation of the equation for long-run elasticities can be found in Appendix C.
28 December 2007
Estimating Air Travel Demand Elasticities
4.4.1
Page 24
Average Fare
Average fare price was used to measure the price of air travel. Although there are other
components to air travelers total cost of travel (taxes, fees, costs of commuting to/from airports,
etc.), fare prices provide a reasonable indication of own-price elasticities – as seen extensively in
prior literature.
Fare prices used in the regression analysis reflect average route fares over the period reported. In
the cases where data needed to be aggregated, a weighted average fare was calculated based on
traffic levels. Average fare appears in all model specifications.
4.4.2
Gross Domestic Product
Gross domestic product (GDP) was used to measure the effect of income on air travel. In general,
an increase in income will make air travel more affordable and therefore increase traffic. GDP
estimates are widely available, and thus provide a variable that can be consistently defined
between regions and over time.
InterVISTAS experimented with the use of several forms of GDP to obtain the variable that
provided high levels of explanatory power and was an economically intuitive form. Within the U.S.,
GDP estimates are available at the city pair level, and a geometric mean is used as the
explanatory variable in the domestic U.S. regression analysis.18 Regression analysis using the IPS
data set used UK national GDP as the explanatory variable, as the dataset was exclusive to
outbound passengers. Regression analysis for all other regions used the geometric mean of GDP
at the national level. GDP estimates were obtained from the World Bank, which converted the
levels from local currency to U.S. dollars at annual average market rates. Although a smaller level
of aggregation would be preferred, the costs to obtain GDP estimates at such a level would be
quite large. GDP appears in all model specifications.
It should be noted that although many of the international models produced low or even negative
income elasticities, this should not necessarily be interpreted as the population’s air travel
sensitivity to changes in income. Due to data limitations of GDP only appearing at the national
level, there may not be adequate variation in the figure to produce reasonable results. While the
income elasticities given for the U.S. domestic market and the UK outbound market can be
interpreted as relatively robust estimates, InterVISTAS cautions the use of results found in other
markets.
4.4.3
Population
In many models, population is used as an explanatory variable in regression analysis. Population
has a direct effect on the size of a market and if omitted from the model specification, may cause a
bias in the estimates. For example, a large increase in traffic may reflect a sudden boom in
population rather than a change in the sensitivity to price or other effects. If population is omitted
from the model, a bias will occur in the estimates of the other variable coefficients. The geometric
18
The geometric mean of GDP is defined as the square root of the GDP of the origin region multiplied by the GDP of the
destination region:
28 December 2007
OriginGDP × DestinationGDP
Estimating Air Travel Demand Elasticities
Page 25
mean of population at the city level was used where possible and at the national level elsewhere.
Population was tested in all model specifications for statistically significant and reasonable results.
The best results tended to be at the city pair level, and this parameter therefore only appears in
U.S. domestic regressions (where city pair population figures are easily obtainable for all
observations).
4.4.4
Route Distance (Trip Length)
The use of route distance was used explored as both an explanatory variable in OLS regression
and as an instrumental variable in 2SLS regression.
The intuition behind the use of distance as an explanatory variable reside in its ability to address
value of travel time savings and availability of substitutes. That is, as haul length increases, the
travel time savings over other modes becomes progressively larger and the viability of other modes
to circumvent obstacles that may arise (such as oceans, poor terrain, unfriendly states, etc.)
becomes progressively smaller. InterVISTAS experimented with the use of different forms of
distance including the form taken in gravity models where travel is defined as a function of the
inverse square of distance; none of the different forms were found to cause a change in
conclusions, therefore the easily interpretable Cobb-Douglas formulation was used.
The use of route distance as an instrumental variable in 2SLS requires a completely different set of
assumptions about its effect on air travel. 2SLS requires that distance be exogenous to the model
(uncorrelated with traffic). Although there are many situations where this will not be the case,
domestic travel in the continental U.S. may be a reasonable market to assert this assumption.
There are fewer travel obstacles within a country with extensive infrastructure than there are
between countries with less infrastructure. Also, once airport commute and check in times are
added to in flight time, the time savings difference between the shortest routes in the U.S. and the
longest become less significant. The use of distance as an IV is also supported by correlation
figures obtained: the correlation between distance and traffic is relatively small at -0.086, while the
correlation between distance and fare is 0.552.
4.4.5
Substitute Goods
InterVISTAS conducted regression analysis on a subset of routes using substitute goods as
explanatory variables. The theory behind this specification is as follows: The demand for a good is
influenced by own price elasticities and cross price elasticities. In the case of air travel, own price
elasticity refers to the sensitivity to changes in fares on a given route, while the cross price
elasticity would be any substitute for travel on the given route. In these specifications, a route
substitutes could be defined as a different airport serving the same catchment area (e.g. ChicagoO’Hare versus Chicago-Midway) or a different destination serving the same purpose (e.g. Las
Vegas versus Reno both serving as recreation destinations).
When variables that should be included as explanatory variables are omitted from the model
employed, a bias in the estimates of the other coefficients will occur under most circumstances.19
In the case of air travel, the exclusion of a proper substitute will bias the fare elasticity estimates
19
A bias will occur in all cases except when the omitted variable is uncorrelated with the included explanatory variables.
28 December 2007
Estimating Air Travel Demand Elasticities
Page 26
obtained in regression analysis. The extent of this bias in a simple linear regression of traffic onto
fare will be equal to the coefficient of the substitute (the cross price elasticity of the substitute)
multiplied by the correlation between fare and the fare substitute.20
Since the coefficient on the missing substitute variable is known to be positive (an increase in the
price of the substitute fare will tend to increase the traffic on the target route) and the correlation
between a fare and a substitute fare is positive (a decrease in the price of a substitute fare will tend
be accompanied by a decrease in the price of the targeted fare) the bias of the fare elasticity
estimate is known to be positive. This implies that a fare elasticity estimate that excludes a
meaningful substitute in the regression model will produce more inelastic estimates than a correctly
specified model would produce. That is, the fare elasticities produced will have an inelastic bias.
InterVISTAS experimented with the use of fares on similar routes as the value of a substitute good
in a small subset of the DB1B data set.21 The results were consistent with what the theory
described above would suggest; the inclusion of a substitute fare was found to increase fare
elasticity estimates.
To produce estimates that would apply to the full data set would require careful selection of proper
substitutes for each route examined and is beyond the scope of this project. However, the general
conclusions supported by theory and the results from the subset analysis provide an indication that
previous estimates will have a bias towards inelastic results.
4.4.6
Real Exchange Rates
InterVISTAS experimented with the use of real exchange rates as an explanatory variable in
determining the demand for leisure air travel.22 The real exchange rate is defined as the consumer
price index (CPI) in the foreign country divided by the CPI in the domestic country divided by the
exchange rate (foreign currency per unit of domestic currency). The intuition behind the use of this
variable is that as the foreign country becomes more expensive (inexpensive), leisure travelers will
travel to the foreign country less (more).
InterVISTAS was unable to obtain robust estimates using this variable; this is likely attributable to a
fundamental error in the variable’s original design. By using CPIs as units to determine relative
costs of travel, there is an implicit assumption that changes in the cost of tourism are equal to
changes in the average cost of goods in the country, and this is not necessarily the case. For
example, a country with 10% inflation in the general economy, but zero inflation in the tourism
sector would have a high real exchange rate while there would be little effect on the cost of travel
to that country.23 Real exchange rates were excluded from final model specifications.
20
A proof of this statement can be found in Appendix B.
21
An example of a substitute good in the data subset would be the use of fares from New York to Fort Lauderdale serving as a
substitute for travel from New York to Miami.
22
Real exchange rates have also been referred to as the ‘Effective Price of Tourism’ in other studies.
23
Assuming no change in nominal exchange rates.
28 December 2007
Estimating Air Travel Demand Elasticities
4.4.7
Page 27
Time Variables
The use of several different forms of time variables was explored. However, only the introduction
of quarterly (seasonal) time dummies was found to consistently increase goodness of fit. The use
of quarterly dummies also tended to increase fare elasticities. This suggests that quarterly factors
are present in both fares and innate traffic demand. These results suggest potential for future
research to model unique quarterly factors for each O/D pair or for types of routes. As such a
specification would be complex and coding exceedingly laborious, the specification was excluded
from InterVISTAS’ analysis and this report.
4.4.8
Route Dummies
The introduction of route dummies in the regressions was found to greatly improve the goodness of
fit. That is, fare, population and income alone do not explain much of the difference in traffic levels
on O/D pairs. Other factors are important. E.g., whether or not non-stop service is available,
whether there are competing carriers operating on non-stop routes, whether there is a low cost
carrier, what types of cultural or financial linkages there are between pairs of cities, etc., appear to
explain much of the differences in traffic levels. By utilizing route dummy variables, these issues
can be controlled without the need to spend extensive resources quantifying them for regression
analysis. Furthermore, in the absence of variables capturing the substitute option (crosselasticities), the dummies can control for the cross-elasticity effects to some degree although
provide no information on the values of these cross-elasticities.
The introduction of route dummies was generally found to make fare elasticities more elastic (i.e.,
makes the elasticities more negative), consistent with the idea that the dummies capture crosselasticity effects.
4.5
Econometric Results
The following section provides an overview of the results obtained through InterVISTAS’
econometric analysis. Elasticity estimates are summarised in Figures 4-1, 4-2, and 4-3.
4.5.1
U.S. DOT Database 1B
Econometric analysis of the U.S. Department of Transport database (DB1B) benefitted from the
large volume of research already conducted on data set in previous studies, and also produced the
most robust results. Final model specifications used real average fares, geometric mean of city
pair real income, distance (explanatory variable in OLS, instrumental variable in 2SLS), time
dummies for every quarter, and route dummies.24 25
OLS and 2SLS regression techniques were both used to evaluate the DB1B data set. ARDL
regression techniques were not used on this data set as the U.S. domestic market capacity
adjustments are deemed to be more responsive to changes in demand than other markets. Price
24
‘Real average fares’ denote average nominal fares that are adjusted for inflation.
25
Geometric mean is defined as the square root of the product of the two figures. In the case of city pair incomes, the geometric
mean would be:
IncomeCity 1 × IncomeCity 2
28 December 2007
Estimating Air Travel Demand Elasticities
Page 28
elasticity estimates were found in the range of -0.8 to -1.5 (mildly inelastic to elastic), income
elasticity estimates were between 1.0 and 1.8 (elastic), and population elasticity estimates were
between 0.5 and 1.8 (inelastic to elastic). Variation in the elasticity estimates is explained in
sections 4.5.4 and 4.5.5..
4.5.2
International Passenger Survey Database
In order to ensure reasonable estimates were produced, this study first sought to replicate the
results of the UK Civil Aviation Authority (CAA) report on demand for outbound air travel using the
International Passenger Survey (IPS) database.26 The model prescribed by the CAA was then
adjusted by InterVISTAS to provide insight on regional and aggregation issues. In reproducing the
results of the CAA, InterVISTAS found that the ‘Effective Price of Tourism’ variable implemented in
the CAA study did not produce robust estimates and also created multicollinearity issues with the
GDP (income) variable. This variable was therefore dropped in InterVISTAS’ final model
specifications.27 Final model specifications used nominal average fares, real GDP, distance,
seasonal dummies, and route dummies.28
Analysis using OLS, 2SLS, and ARDL methods was attempted in the IPS data set, however 2SLS
results were rejected by InterVISTAS due to poor performance by the instrumental variable. Price
elasticity estimates in the IPS database were found in the range of -0.8 to -0.9 (mildly inelastic)
while income elasticity estimates were between 1.2 and 1.5 (mildly elastic to elastic). Population
variables did not yield statistically significant results and were therefore dropped from the final
model specification.
4.5.3
PaxIS Database
PaxIS database analysis differed from DB1B and IPS analysis due to the relatively short time
series available (data begins in 2005). However, the global nature of the dataset enabled a wider
range of cross-sectional data analysis. Final model specifications of regional subsets used
nominal average fares, geometric mean of real GDP, distance, and time dummies for each month
of data. Final model specifications of the full (world) data set include geometric mean of
population, in addition to the above variables.
Analysis using OLS, 2SLS, and ARDL methods was conducted with the PaxIS data set, however
2SLS results were rejected due to poor instrumental variables. Price elasticities between -0.4 and
-2.9 were calculated, with significant variation between regions, and between OLS and ARDL
models.29 Income and population elasticities were determined to be weak estimates since input
figures were only available at the national level, not the city pair level. It is noted that national
income/population levels do not likely reflect the air catchment area levels under these
circumstances.
26
Civil Aviation Authority (2005), “Demand for Outbound Leisure Air Travel and its Key Drivers”
27
Other issues with the ‘Effective Price of Tourism’ (real exchange rate) variable are described in section 4.4.6.
28
Nominal fares were used as the CAA study did not note any adjustment of IPS fare data for inflation.
29
Regional variations are described in section 4.5.6.
28 December 2007
Estimating Air Travel Demand Elasticities
4.5.4
Page 29
Route Level Analysis
The outcomes from InterVISTAS’ econometric analysis supports a route level elasticity near -1.4.
This finding is likely the result of substitutability between various routes for one another. The
results from analysis using route dummies (which may be capturing all or part of the effect of the
price of substitutes) and experiments on a sample set of routes with price of substitutes support
this finding.
Regressions on the U.S. DB1B data set using 2SLS are especially revealing in this regard.
However, as discussed in section 4.3.2, the 2SLS technique is difficult to implement with a limited
budget and with global data, as it requires the assembly of one or more instruments from the
supply equation. Using distance as an instrument produced results that further supported an
elasticity near this level. Nevertheless, there still is some concern over distance’s use as an
instrument, due to its perceived exogenous influence on demand.
4.5.5
Aggregate Level Analysis
InterVISTAS’ econometric analysis supports the finding that elasticities (associated with price
changes) at more aggregated levels (E.g. National/Pan-National) should be substantially more
inelastic than route level changes. Based on the results seen in all three data sets, a reasonable
range of fare elasticities would be from -0.6 to -0.85, with the more inelastic end of the range
describing pan-national effects and the more elastic end of the range describing national effects.
Inelastic results were found over a range of model specifications which excluded route dummies.
This elasticity range is essentially a combination of the route own price elasticity with cross price
elasticities, when all routes have prices that vary identically. The less elastic result is consistent
with observations that part of the price elasticity observed from low cost carriers (LCCs) at
secondary airports involves diversion from primary airports in the catchment area, or from diversion
from trips on other routes. When this is controlled for, LCCs have a lower level of market
stimulation, consistent with less elastic national elasticities.
4.5.6
Geographic Market Analysis
Elasticity estimates were found by InterVISTAS to vary by a statistically significant level in different
geographic regions. This section describes the regional socio-economic characteristics that affect
key travel markets and provides an overview of findings from InterVISTAS’ econometric analysis:30
Intra North America: Elasticity estimates were broadly consistent with what was found in the
literature review. The market is well established with relatively high levels of capacity and traffic.
Fares tend to be low, while haul lengths are short to medium.
Intra Europe: Higher elasticities were observed in the Intra Europe market. Intra Europe flights
are noted by shorter average distances and use of very low fares that result in significant market
stimulation. The low fares in Europe are not observed in North America, this is consistent with the
30
Elasticity comparisons are noted to be relative to results observed in Intra North America analysis.
28 December 2007
Estimating Air Travel Demand Elasticities
Page 30
higher elasticities observed in Europe. In the past, the European market had high charter carrier
share, which today is merely being converted to very low fare LCCs.
Intra Asia: Moderately more inelastic estimates were found in Intra Asia. The LCC phenomena is
emerging in Asia, but a modest sized middle class in many markets suggests somewhat weaker
non-business demand levels and less elastic demand. This provides context for the estimates
obtained in the region.
Intra Sub-Saharan Africa: This market supports a relatively inelastic demand estimate as
compared to North America. These economies have a small middle class population, resulting in a
high weight of traffic being driven by higher income individuals, which are less elastic.
Intra South America: Estimates obtained from regression analysis produced highly elastic fare
elasticities, however this result may be anomalous. There is an emerging middle class which
makes the market more elastic than sub-Saharan Africa, and LCCs are emerging in Brazil, Chile,
and Mexico, but econometric findings suggested that this market has highest elasticity in the world.
It is suggested that use some caution be used with Intra South America elasticity estimates.
Trans Atlantic (North America – Europe): A high price elasticity was found in the Trans Atlantic
market. This market is often observed to have fares only slightly higher than domestic US fares,
consistent with high price elasticity. The market has also been well developed by charter carriers,
which is also consistent with high price elasticity. Price is likely more important than frequency in
this market than in Intra North America.
Trans Pacific (North America – Asia): A lower price elasticity was observed in the Trans Pacific
travel market. Travel in the market has had no charter services, and continues to have major
markets with less liberal pricing provisions (Japan, China). There has been some recent
emergence of long haul LCCs (e.g., Oasis) but the present environment supports a market less
elastic than Intra North America (since long haul traffic tends to be less elastic) and less elastic
than trans Atlantic (since trans Atlantic serves a substantial middle class).
Europe-Asia: A moderately inelastic price elasticity was found in the market for travel between
Europe and South/South-East Asia (Europe-Asia). This result is in contrast to the results found in
the respective intra markets of Europe and Asia, and provides further evidence for lower elasticities
on long-haul and intercontinental air transportation.
4.6
Econometric Results
This section provides the elasticity estimates from the final model specifications. Full regression
outputs are provided in Appendix D. It should be noted that the most robust results were obtained
for the U.S. Domestic market. These figures can therefore be interpreted as the most reliable.
Figures for other regions provide a base of reference, but should be jointly interpreted with
conclusions from the literature and key results from the U.S. regression analysis. Figure 4-1
provides elasticity estimates for the final OLS regression equations. Figure 4-2 presents the
results of 2SLS regression analysis on the U.S. domestic market. 2SLS results for other regions
are not reported as no suitable IV is available. Figure 4-3 presents the results of ARDL regression
analysis.
28 December 2007
Estimating Air Travel Demand Elasticities
Page 31
Figure 4-1: Ordinary Least Squares Regression Results
Region
Price
Elasticity
Income
Elasticity
Population
Elasticity
Goodness of
Fit
U.S. Domestic
-0.83
1.00
0.45
0.95
World
-0.62
0.03
0.10
0.61
UK to Western Europe
-0.82
1.45
N/A
0.61
Intra-Europe
-1.30
0.09
N/A
0.71
Intra South and East Asia
-0.86
-0.08
N/A
0.10
Intra Latin America
-1.20
0.10
N/A
0.22
Intra Sub-Saharan Africa
-0.56
0.19
N/A
0.60
Transatlantic
-1.06
0.32
N/A
0.60
Transpacific
-0.36
0.19
N/A
0.60
Europe-Asia
-0.73
0.22
N/A
0.40
Figure 4-2: Two-Stage Least Squares Regression Results
Region
U.S. Domestic
Price
Elasticity
Income
Elasticity*
Population
Elasticity
Goodness of
Fit**
-1.46
1.76
1.80
N/A
*See Section 4.4.2 (Gross Domestic Product) for information on interpreting income elasticities.
**The R-Squared value is excluded since 2SLS seeks to obtain proper parameter estimates, not maximize goodness of
fit (unlike OLS).
Figure 4-3: Autoregressive Distributed Lag (OLS) Regression Results
Region
Price
Elasticity
Income
Elasticity*
Population
Elasticity
Goodness of
Fit
U.S. Domestic
N/A
N/A
N/A
N/A
World
-2.03
-0.34
N/A
0.88
UK to Western Europe
-0.86
1.18
N/A
0.70
Intra Europe
-2.78
-0.28
N/A
0.87
Intra South and East Asia
-2.90
-1.48
N/A
0.88
Intra Latin America
-1.80
-0.43
N/A
0.86
Intra Sub-Saharan Africa
-1.33
0.06
N/A
0.61
Transatlantic
-1.42
-0.73
N/A
0.88
Transpacific
-1.08
-2.2
N/A
0.84
Europe-Asia
-1.24
0.23
N/A
0.86
*See Section 4.4.2 (Gross Domestic Product) for information on interpreting income elasticities.
28 December 2007
Estimating Air Travel Demand Elasticities
Page 32
5.0 Conclusion: Synthesising the Findings
5.1
Introduction
The literature review and econometric analysis demonstrated that air fares elasticity varies
depending on a number of factors such as geography, distance and level of aggregation.
Ultimately, the determining the right elasticity value to use depends on the type of question being
asked. Examining the traffic impact of a fare increase on a given route requires a different
elasticity than when examining the impact of an across-the-broad fare increase on all routes in a
country or region.
This chapter attempts to synthesise the results of the literature review and econometric analysis to
develop some general guidelines on the use and application of air fare elasticities. While the
literature review and econometrics did not provide a full set of elasticities covering all possibilities,
they did provide sufficient information to infer a full set of elasticities. In particular, the following
characteristics were identified as affecting the elasticity values, as discussed below.
Level of Aggregation
As noted previously, the level of aggregation affects the elasticity value:
ƒ
ƒ
ƒ
ƒ
ƒ
Fare-class Level: this the most disaggregate level, where travellers are choosing different fare
classes on particularly airlines. At this level, the elasticities are arguably highest, as travellers
can easily switch between fare-class levels.
Carrier Level: the literature review found that the fare elasticity facing a particular carrier on a
given route can be quite high (e.g., -1.5 to -2.5). If an air carrier increases it fare unilaterally, it
is likely to lose passenger to other carriers operating on that route.
Route/Market Level: at the route or market level, the elasticity response is generally lower, as
evidenced by both the literature review and the econometric analysis. Travellers faced with a
fare increase on all carriers serving the route (e.g., due to an increase in airport fees and
charges), have the option of not travelling, travelling to another destination, traveling on
another route (e.g., LGW-CDG rather than LHR-CDG) or travelling by another mode.
National Level: at the national level, fare elasticities would be expected to be lower, as
travellers have less options for avoiding the fare increase. For example, if a national
government imposed a new or increased tax on aviation, traveller could only avoid this
increase by using another mode (which may not always be possible) or not travelling (or
travelling elsewhere).
Pan-National Level: at the pan-national level (e.g., the European Union), the options for
avoiding the fare increase are even further reduced, so therefore the elasticity would be
expected to be lower.
Given the economic and policy focus of this study, the first two levels of aggregation are outside
the scope of this study. Nevertheless, elasticity values for these levels of aggregation can be
found in the literature review in Chapter 3.
28 December 2007
Estimating Air Travel Demand Elasticities
Page 33
Length of Haul
The literature review consistently found that the fare elasticities on short-haul routes were generally
higher than on long-haul routes. In part, this reflects the opportunity for inter-modal substitution on
short haul routes (e.g., travellers can switch to rail or car in response to air fare increases).
Therefore, in this study a differentiation has been made between short-haul and long-haul
elasticities.
Geographic Aviation Market
The characteristics of the market or region being examined can also affect the fare elasticity used.
This reflect a number of a number of factors such as economic development, aviation market
structure, government regulation (e.g., regulated vs liberalised), demographic factors, historical
factors, etc. In develop the guideline elasticities, the following major geographic markets have
been identified:
ƒ
ƒ
ƒ
ƒ
ƒ
ƒ
ƒ
ƒ
Intra North America
Intra Europe
Intra Asia
Intra Sub-Sahara Africa
Intra South America
Trans Atlantic (North America – Europe)
Trans Pacific (North America – Asia)
Europe – Asia
5.2
General Guidelines on Air Fare Elasticity Values
The following elasticities have been developed as a synthesis of the literature review and the
econometric analysis. The approach taken was to develop three base elasticities reflecting the
levels of aggregation (route, national and pan-national level). Multiplicative adjustors were then
developed to adjust the elasticities to reflect specific markets.
Base Elasticities
The base elasticities were developed from the analysis of the various datasets and the generalised
findings from the literature review.
ƒ
ƒ
Route/Market Level: -1.4
The literature review found that elasticities at the route or market level in the range of -1.2 to 1.5. This was verified by our own econometric analysis of the U.S. DB1A where it was
possible to capture the effects of route substitution through the use of route dummy variables
and experiments using variables capturing the price of route substitutes. These regressions
produced fare elasticities in the region of -1.4, which has been used as the base elasticity at
the route level.
National Level: - 0.8
The econometric analysis of all three datasets found that without the route substitution term,
the analysis produced elasticities in the region of -0.8. This elasticity is essentially a
combination of the route own price elasticity with cross price elasticities when all national
routes have prices which vary identically. Thus, less elastic result is consistent with
observations that part of the so called price elasticity observed from LCCs at secondary
28 December 2007
Estimating Air Travel Demand Elasticities
ƒ
Page 34
airports involves diversion from primary airports in the catchment area, or from diversion from
trips on other routes. When this is controlled for, LCCs have a lower level of market
stimulation, consistent with less elastic national elasticities.
Pan-National Level: -0.6
This result is conceptually derived from the same mathematics which show the relationship
between route and national elasticities. In this case, a much broader set of routes (e.g., across
a continent) experience an identical price change.
Thus the route elasticity is applicable to a situation where the price of an individual route changes,
for example, the fare on Warsaw-Coventry increases with the price of routes from Warsaw to other
UK and other European points remain unchanged. The national elasticity applies to situation such
as all Warsaw-UK prices changing identically, but price from Warsaw to other European points
being unchanged. Pan national changes apply where prices from Warsaw to all points in Europe
change identically.
Geographic Aviation Market
The econometric analysis of the IATA PaxIS data found considerable differences between aviation
markets. Based on this analysis, the following elasticity multipliers have been developed:
Geographic Market
Elasticity Multiplier
Intra North America
1.00
Most research uses US data. This is our
reference point.
1.40
Shorter average distances, observed use of very
low fares resulting great market stimulation. The
significantly low fares in Europe (relative to North
America) are consistent with higher elasticities in
Europe. Traditionally the European market had
high charter carrier share, which today is merely
being converted to very low fare LCCs.
Intra Asia
0.95
The LCC phenomena is emerging in Asia, but
modest sized middle class in many markets
suggests somewhat less elastic than in North
America
Intra Sub-Sahara Africa
0.60
These economies have limited middle class,
resulting in high weight on higher income
individuals who are less elastic
1.25
There is an emerging middle class which makes
the market more elastic than sub-Saraha Africa,
and LCCs are emerging in Brazil, Chile, and
Mexico.
1.20
This market is often observed to have fares only
slightly higher than domestic U.S. fares,
consistent with high price elasticity. Market has
been well developed by charter carriers,
consistent with high price elasticity. Price is
likely more important than frequency in this
Intra Europe
Intra South America
Trans Atlantic
(North America – Europe)
28 December 2007
Comment
Estimating Air Travel Demand Elasticities
Geographic Market
Page 35
Elasticity Multiplier
Comment
market than in domestic U.S.
Trans Pacific
(North America – Asia)
0.60
TransPacific has had no charter services, and
continues to have major markets (Japan, China)
with less liberal pricing provisions. Some
emergence of long haul LCCs (e.g., Oasis) but at
present this market seems to be less elastic than
domestic US and than the well developed trans
Atlantic which serves a substantial middle class
Europe – Asia
0.90
This market has marginally lower elasticities than
the U.S. domestic market.
Short-Haul/Long-Haul Adjustment
The literature review consistently found that the fare elasticities on short-haul routes were generally
higher than on long-haul routes. In part, this reflects the opportunity for inter-modal substitution on
short haul routes (e.g., travellers can switch to rail or car in response to air fare increases). While
the geographic breakdowns capture some variation by length of haul, there is still considerable
variation within each market. In particular, very short-haul flights (approximately less than 1 hour
flight time) are subject to greater competition from other modes.
On this basis, the following short-haul multiplicative adjustors can be applied to analysis of shorthaul routes:
ƒ
Short haul: 1.1 (+10%).
Note that this adjustor does not apply to analysis of the trans-Atlantic and trans-Pacific market,
which are considered entirely long haul, with virtually no opportunity for modal substitution.
Application of Adjustors
Elasticities for different situations can be developed by selecting the relevant base elasticity and
applying the relevant multipliers, for example:
1.
To examine the impact of an EU-wide aviation tax on short-haul markets, the elasticity
would be developed as follows:
- Base multiplier: -0.6 (pan-national)
- Geographic market: 1.4 (Intra Europe)
- Short-haul adjustor: 1.1
The multiplier would then be calculated as: -0.6 x 1.4 x 1.1 = -0.924.
2.
To examine the impact of a UK tax on aviation on Trans Atlantic traffic, the elasticity
should be developed as follows:
- Base multiplier: -0.8 (national)
- Geographic market: 1.2 (Trans Atlantic)
The multiplier would then be calculated as: -0.8 x 1.2 = -0.96.
28 December 2007
Estimating Air Travel Demand Elasticities
3.
Page 36
To examine the impact of impact of an increase in airport landing fees on a particular
short-haul route in Asia, the elasticity should be developed as follows:
- Base multiplier: -1.4 (route)
- Geographic market: 0.95 (Intra Asia)
- Short-haul adjustor: 1.1
The multiplier would then be calculated as: -1.4 x 0.95 x 1.1 = -1.46.
The full range of possible elasticities is presented in the table below. The route level elasticities
range from -0.84 to -1.96 depending on the geographic market and length of haul. The national
level elasticities range from -0.48 to -1.23, while the pan-national elasticities range from -0.36 to 0.92.
Route/Market Level
National Level
Pan-National Level
Shorthaul
Longhaul
Shorthaul
Longhaul
Shorthaul
Longhaul
Intra North America
-1.54
-1.40
-0.88
-0.80
-0.66
-0.60
Intra Europe
-1.96*
-1.96
-1.23
-1.12
-0.92
-0.84
Intra Asia
-1.46
-1.33
-0.84
-0.76
-0.63
-0.57
Intra Sub-Sahara Africa
-0.92
-0.84
-0.53
-0.48
-0.40
-0.36
Intra South America
-1.93
-1.75
-1.10
-1.00
-0.83
-0.75
Trans Atlantic (North America – Europe)
-1.85
-1.68
-1.06
-0.96
-0.79
-0.72
Trans Pacific (North America – Asia)
-0.92
-0.84
-0.53
-0.48
-0.40
-0.36
Europe-Asia
-1.39
-1.26
-0.79
-0.72
-0.59
-0.54
*The short-haul adjustor has not been applied to the Intra Europe short-haul elasticity in order to maintain elasticities below 2.0
5.3
Comment on Income Elasticities
The focus of this study was on fare elasticities. Nevertheless, the analysis also considered income
elasticities as part of the analytical process, albeit in less detail than fare elasticities. The literature
review found a wide range of income elasticities, generally in the range +1.0 to +2.0. That is, air
travel is generally found to be income elastic.
This is supported by more casual analysis which plots air travel per capita against income per
capita across a wide range of nations, and finds a strong, positive relationship between the two.
That is as income per capita grows, air travel per capita grows, and the growth in air travel is faster
than the growth in income. These types of two dimensional analysis typically show some taper
among the nations with the highest per capita incomes. That is, the income elasticity appears to
decline somewhat at higher incomes, although staying elastic above 1.0.
Our statistical analysis produced mixed results.
ƒ
Analysis with the well developed U.S. data produced income elasticities generally in the range
of +1.6 to +1.8. We do note that the income elasticity results are not robust, in that small
changes in model specification can result in sizeable changes in the income elasticity.
28 December 2007
Estimating Air Travel Demand Elasticities
ƒ
ƒ
ƒ
ƒ
Page 37
Correlation between income and time trends or shifts, and correlation between income and the
use of route specific indicators (i.e., route dummy variables) appear to affect the results.
When experimental analysis was done with the price of travel on substitute routes, we
generally found slightly higher income elasticities. That is, at the “route level”, income elasticity
was higher than at the “national level”.
The UK IPS data generally found slightly lower income elasticities than the U.S., data, with
ranges from +1.2 to +1.6 typically found.
The UK IPS data also revealed higher income elasticities for long haul routes. This would be
consistent with an hypothesis that middle to lower income individuals are apt to travel on short
to medium haul routes, but far less likely to travel on long haul. However, long haul sunspot
travel is not uncommon from the UK, even for lower middle income individuals. The
reconciliation may be that higher income leads to higher frequency of long haul travel.
Results using the global PaxIS database were largely unsuccessful. In part, this was due to
data challenges, as assembling income data on individual cities around the world was not
possible. Outside of the developed world, GDP and income measures are generally not
available at a city level, at least from global information sources which are constructed on a
consistent basis. Analysis thus used national incomes as proxies for city incomes, but this
introduced its own statistical bias. Results showed no stability to model specification and thus
were rejected.
Based on the above findings, we recommend the use of the income elasticities in the tables below.
These were constructed using professional judgement, based on information from the literature
review, new results obtained in this study, and the two-dimensional analysis of travel propensity
versus national income across a wide range of nations.
Route level
Length of Haul
U.S.
Other Developed
Nations
Developing
Economies
Short haul
1.8
1.5
2.0
Medium haul
1.9
1.6
2.0
Long haul
2.0
1.7
2.2
Ultra long haul
2.2
2.4
2.7
28 December 2007
Estimating Air Travel Demand Elasticities
Page 38
National level
Length of Haul
U.S.
Other Developed
Nations
Developing
Economies
Short haul
1.6
1.3
1.8
Medium haul
1.7
1.4
1.8
Long haul
1.8
1.5
2.0
Ultra long haul
2.0
2.2
2.5
5.4
Conclusions
As described in the literature review, the consensus among most aviation economists has been
that demand for airline services is generally both price and income elastic. The latter suggests that
air transport has the characteristics of a luxury good. A natural question is how an income elastic
good can also be price elastic.
This report has revealed that there are different price elasticities associated with different uses.
When consumers are choosing between airlines on a route, or even between destinations for
holidays, conference locations, etc., there is a degree of price elasticity for airline seats. However,
if all competitors on a route, or if a wide range of routes all experience the same proportionate price
increase, the demand for airline services becomes less elastic. As a price increase is extended to
ever larger groups of competing airlines or competing destinations, then the overall demand for air
travel is revealed to be somewhat inelastic.
The implications of this are as follows:
ƒ
ƒ
ƒ
For an airline on a given route, increasing price is likely to result in a more than proportionate
decrease in air travel. Lower fares will greatly stimulate traffic and raise revenues.
(i.e., airline specific fare changes are price elastic).
If all airlines on a given route increase fares by the same amount (e.g., due to the imposition of
passenger based airport fees that are passed on to the consumer), then the decrease in traffic
will be less than the former case.
If all airlines on a wide set of routes increase fares by roughly similar amounts (e.g., due to the
imposition of new market-wide taxes or to the working through of higher fuel or security costs)
then the decrease in traffic may be less or much less than proportional to the increase in fares.
(i.e., general increases in airline fares across a broad range of markets appear to be price
inelastic.)
Thus, the particular elasticity value to be used for analysing price effects in airline markets depends
on the question being asked. The narrower the applicability of a price change, the more elastic the
response. The more general the applicability of a price change (perhaps due to higher costs or
taxes) the less elastic the response.
28 December 2007
Estimating Air Travel Demand Elasticities
Page 39
Appendix A - Final Model Specifications
The following section describes the models used to provide the figures seen in Section 4.0 of the
report. Error terms are excluded from the specifications for ease of reading.
Ordinary Least Squares
U.S. Domestic:
ln(Traffic ) = Constant + a1 × ln(Fare) + a 2 × ln(GeometricCityIncomePerCapita )
+ a 3 × ln(GeometricCityPopulation ) + a 4 × ln(RouteDistance )
m −1
n
i =5
j =m
+ ∑ a i × (TimePeriodDummies ) + ∑ a j × (RouteDummies )
UK to Western Europe:
ln(Traffic ) = Constant + a1 × ln(Fare) + a 2 × ln(UKGDP ) + a 3 × ln(RouteDistance )
6
+ ∑ a i × (QuarterlyDummies )
i =4
Other:
ln(Traffic ) = Constant + a1 × ln(Fare) + a 2 × ln(GeometricGDPPerCapita )
+ a 3 × ln(GeometricCountryPopulation ) + a 4 × ln(RouteDistance )
n
+ ∑ a i × (TimePeriodDummies )
i =5
Two Stage Least Squares
U.S. Domestic:
ln(RouteDistance) used as instrument in analysis.
ln(Traffic ) = Constant + a1 × ln(Fare) + a 2 × ln(GeometricCityIncomePerCapita)
m −1
n
i =4
j =m
+ a 3 × ln(GeometricCityPopulation) + ∑ a i × (TimePeriodDummies ) + ∑ a j × (RouteDummies )
Autoregressive Distributed Lag
UK to Western Europe:
ln(Traffic t ) = Constant + a1 × ln(Traffic t −1 ) + a 2 × ln(Fare t ) + a 3 × ln(UKGDPt )
7
+ a 4 × ln(RouteDistance) + ∑ a i × (QuarterlyDummies )
i =5
Where ‘t’ is the period of the current observation and ‘t-1’ is the period of the observation in the
same quarter/month of the previous year.
Other Non-U.S. Domestic:
28 December 2007
Estimating Air Travel Demand Elasticities
Page 40
ln(Traffic t ) = Constant + a1 × ln(Traffic t −1 ) + a 2 × ln(Fare t ) + a 3 × ln(GeometricGDPPerCapita )
n
+ a 4 × ln(RouteDistance) + ∑ a i × (TimePeriodDummies )
i =5
Where ‘t’ is the period of the current observation and ‘t-1’ is the period of the observation in the
same quarter/month of the previous year.
28 December 2007
Estimating Air Travel Demand Elasticities
Page 41
Appendix B - Omitted Variable Bias
The following section provides a brief proof of the bias that occurs in fare elasticity estimates when
the price of a substitute good is excluded from the regression model.
Suppose the true model has a specification:
y = β 0 + β1x1 + β 2 x 2
Where:
y = Traffic
x 1 = Average Fare
x 2 = The Price of a Substitute
e = An Error Term
An unbiased regression equation would be:
yˆ = βˆ 0 + βˆ1 x 1 + βˆ 2 x 2 + e
If x 2 is omitted from the regression, then the following ‘short regression’ equation is estimated:
~
~
y~ = β 0 + β 1 x 1 + u
It is useful to know that:31
~
~
β 1 = βˆ1 + βˆ 2δ 1
~
Where δ is the sample covariance between x 1 and x 2 divided by the sample variance of x 1 .
The bias is therefore:
~
~
~
E ( β 1 ) − β 1 = E ( βˆ1 + βˆ 2δ 1 ) − β 1 = β 2δ 1
~
Since δ 1 is interpreted as being positive since the route fare and the substitute fare face the same
economic and competitive pressures and β 2 is known to be positive by definition of a substitute,
then:
~
β 2δ 1 > 0
That is, fare elasticity estimates have a positive bias.
31
Source: “Introductory Econometrics: A Modern Approach”, Jeffrey M. Woolridge.
28 December 2007
Estimating Air Travel Demand Elasticities
Page 42
Appendix C - Long-Run Elasticity in ARDL Model
Since traffic appears on both sides of the equation in ARDL models described in Appendix A, the
coefficients on the explanatory variables cannot be directly interpreted as long-run elasticities (only
short-run elasticities). The long-run elasticities are defined as when traffic across time periods
stabilize.
That is, if the ARDL model is defined as an equation with a single lag in the dependent variable
and one contemporary independent variable:
ln( y t ) = β 0 + β 1 ln( y t −1 ) + β 2 ln( x t ) + e t
Where ‘t’ is the period of the current observation and ‘t-1’ is the period of the observation in the
same quarter/month of the previous year. Long-run elasticities are defined when:
y t = y t −1
This implies that:
ln( y t ) = β 0 + β1 ln( y t ) + β 2 ln( x t ) + et
and therefore,
ln( y t ) =
β 0 + β 2 ln( x t ) + e t
1 − β1
If derivatives are taken on both sides with respect to ln( x t ) , the equality becomes:32
∂ ln( y t )
β2
=
∂ ln( x t ) 1 − β 1
Which defines the long-run elasticity.
32
∂ s are used instead of ds to reflect the fact that other explanatory variables may be in the final model specification.
28 December 2007
Estimating Air Travel Demand Elasticities
Page 43
Appendix D - Regression Output
Ordinary Least Squares
Model
Region
Ln Fare
Ln Distance
Ln Income
Ln Population
Time Dummy
Route Dummy
Adjusted R2
Model 1
U.S. Domestic
(DB1B)
-0.83
(-141.03)
-1.73
(-31.85)
0.99
(22.09)
0.45
(8.76)
For every
quarter
Yes
0.95
Model 2
World
(PaxIS)
-0.62
(-19.54)
-0.13
(-6.72)
0.03
(2.64)
0.10
(7.30)
For every
month
No
0.61
Model 3
UK to Western
Europe (IPS)
-0.82
(-36.00)
0.87
(47.56)
1.45
(14.33)
N/A
For every
quarter
No
0.61
Model 4
Intra-Europe
(PaxIS)
-1.30
(-140.73)
0.13
(5.29)
0.09
(14.70)
N/A
For every
month
No
0.71
Model 5
Intra South and East
Asia (PaxIS)
-0.86
(-28.04)
0.13
(5.29)
-0.08
(-5.37)
N/A
For every
month
No
0.10
Model 6
Intra Latin America
(PaxIS)
-1.20
(-52.07)
-0.15
(-11.19)
0.10
(5.05)
N/A
For every
month
No
0.22
Model 7
Intra Sub-Saharan
Africa (PaxIS)
-0.56
(-7.433)
0.12
(2.34)
0.19
(7.99)
N/A
For every
month
No
0.60
Model 8
Transatlantic
(PaxIS)
-1.06
(-35.42)
-0.09
(-2.14)
0.32
(13.53)
N/A
For every
month
No
0.60
Model 9
Transpacific
(PaxIS)
-0.36
(-7.54)
-1.78
(-10.90)
0.19
(6.69)
N/A
For every
month
No
0.60
Model 10
Europe-Asia
(PaxIS)
-0.73
(-18.28)
0.25
(0.43)
0.22
(11.26)
N/A
For every
month
No
0.40
All figures reported are valid at the 95% significance level. Figures in brackets are t-statistics.
28 December 2007
Estimating Air Travel Demand Elasticities
Page 44
Regression Output – Two Stage Least Squares
Model
Number
Region
Ln Fare
Ln Distance
Ln Income
Ln
Population
Time
Dummy
Route
Dummy
Adjusted RSquared
Model 1
U.S.
Domestic
(DB1B)
-1.46
(-46.89)
N/A
1.76
(24.34)
1.80
(35.41)
For every
quarter
Yes
N/A
Notes:
All figures reported are valid at the 95% significance level. Figures in brackets are t-statistics.
Ln Distance was used as the instrumental variable in regression
The R-Squared value is excluded since 2SLS seeks to obtain proper parameter estimates, not maximize goodness of fit (unlike OLS).
28 December 2007
Estimating Air Travel Demand Elasticities
Page 45
Regression Output – Autoregressive Distributed Lag
Model
Number
Region
Year over Year
Lag Ln Traffic
Ln Fare
Ln Distance
Ln Income
Time Dummy
Route
Dummy
Adjusted RSquared
Model 1
World
(PaxIS)
0.92
(1392.64)
-0.15
(-72.56)
-0.06
(-45.94)
-0.03
(-32.38)
For every
month
No
0.88
Model 2
UK to Western
Europe (IPS)
0.80
(120.02)
-0.17
(-8.31)
0.19
(12.01
0.43
(4.53)
Quarterly
No
0.70
Model 3
Intra-Europe
(PaxIS)
0.903
(810.80)
-0.27
(-70.99)
0.04
(17.72)
-0.03
(-11.53)
For every
month
No
0.87
Model 4
Intra South and
East Asia
(PaxIS)
0.95
(324.34)
-0.16
(-13.98)
0.11
(12.35)
-0.08
(-15.49)
For every
month
No
0.88
Model 5
Intra Latin
America
(PaxIS)
0.90
(325.37)
-0.19
(-18.16)
0.06
(10.08)
-0.04
(-5.41)
For every
month
No
0.86
Model 6
Intra SubSaharan Africa
(PaxIS)
0.72
(76.80)
-0.37
(-7.61)
0.23
(7.17)
0.02
(1.11)
For every
month
No
0.61
Model 7
Transatlantic
(PaxIS)
0.93
(427.52)
-0.10
(-8.99)
0.02
(1.36)
-0.05
(-5.92)
For every
month
No
0.88
Model 8
Transpacific
(PaxIS)
0.91
(141.22)
-0.09
(-4.72)
0.21
(2.99)
-0.19
(-16.22)
For every
month
No
0.84
Model 9
Europe-Asia
(PaxIS)
0.92
(261.66)
-0.10
(-6.10)
-0.14
(-5.99)
0.02
(2.35)
For every
month
No
0.86
All figures reported are valid at the 95% significance level. Figures in brackets are t-statistics.
28 December 2007
Estimating Air Travel Demand Elasticities
Page 46
Appendix E - Data Sources
U.S. Domestic Market Data Sources
Variable Name
Source
Units
Time
Frequency
Explanation
Traffic
U.S. DOT DB1B
Persons
Quarterly
Origin-destination
passengers per quarter for a
city pair route.
Real Average
Fare
U.S. DOT DB1B
and U.S. Bureau
of Labor
1994
USD
Quarterly
Calculated by taking
reported average fare from
DB1B 10% sample then
dividing by CPI.
Real MSA
Personal Income
Per Capita
U.S. BEA and U.S
Bureau of Labor
1994
USD
Annually
Calculated by taking
personal income for a given
Metro Statistical Area then
dividing by population and
then dividing by the CPI.
MSA Population
Bureau of Labor
Persons
Annually
Number of residency in the
given Metro Statistical Area.
Real Fuel Price
U.S. EIA and
Bureau of Labor
1994
U.S.
cents
per U.S.
Gallon
Quarterly
Calculated by taking the
price of fuel and dividing by
the CPI.
Distance
U.S. DOT DB1B
Miles
Constant
Actual distance travelled by
route.
28 December 2007
Estimating Air Travel Demand Elasticities
Page 47
UK Outbound Market Data Sources
Variable Name
Source
Units
Time
Frequency
Explanation
Traffic
UK International
Passenger Survey
Persons
Quarterly
Outbound passengers per
quarter for a route.
Separated by business and
leisure.
Average Fare
UK International
Passenger Survey
GBP
Quarterly
Reported average fare from
survey.
Real Gross
Domestic Product
Office for National
Statistics
GBP
Quarterly
UK Real Gross Domestic
Product (Not seasonally
adjusted).
Distance
Great Circle
Mapper
Km
Constant
Airport pair distance based
on great circle.
28 December 2007
Estimating Air Travel Demand Elasticities
Page 48
Appendix F - List of Countries by Region
Africa : Central/Western
Africa
Benin
Burkina Faso
Cameroon
Cape Verde
Central African Republic
Chad
Congo
Congo Democratic
Republic of
Cote D'Ivoire
Equatorial Guinea
Gabon
Gambia
Ghana
Guinea
Guinea-Bissau
Liberia
Mali
Mauritania
Mayotte
Niger
Nigeria
Sao Tome and Principe
Senegal
Sierra Leone
Togo
Africa : Eastern Africa
Burundi
Comoros
Djibouti
Eritrea
Ethiopia
Kenya
Madagascar
Mauritius
Rwanda
Seychelles
Somalia
28 December 2007
Tanzania United Republic
of
Uganda
Africa : North Africa
Mongolia
Russian Federation
Asia : South Asia
Algeria
Egypt
Libyan Arab Jamahiriya
Morocco
Sudan
Tunisia
Afghanistan
Bangladesh
India
Maldives
Nepal
Pakistan
Sri Lanka
Africa : Southern Africa
Asia : South East Asia
Angola
Botswana
Lesotho
Malawi
Mozambique
Namibia
South Africa
Swaziland
Zambia
Zimbabwe
Brunei Darussalam
Cambodia
Indonesia
Lao People's Democratic
Republic
Malaysia
Myanmar
Philippines
Singapore
Thailand
Timor-leste
Viet Nam
Asia : Central Asia
Bhutan
Kazakhstan
Kyrgyzstan
Tajikistan
Turkmenistan
Uzbekistan
Asia : North East Asia
China
Hong Kong (sar) China
Japan
Korea Democratic People's
Republic of
Korea Republic of
Macao (sar) China
Europe : Eastern/Central
Europe
Albania
Armenia
Azerbaijan
Belarus
Bosnia and Herzegovina
Bulgaria
Croatia
Czech Republic
Estonia
Georgia
Hungary
Latvia
Lithuania
Estimating Air Travel Demand Elasticities
Macedonia Former
Yugoslav Republic of
Moldova Republic of
Poland
Romania
Russian Federation
Serbia and Montenegro
Slovakia
Slovenia
Ukraine
Europe : Western Europe
Austria
Belgium
Cyprus
Denmark
Faroe Islands
Finland
France
Germany
Greece
Iceland
Ireland Republic of
Italy
Luxembourg
Malta
Monaco
Netherlands
Norway
Portugal
Spain
Sweden
Switzerland
Turkey
United Kingdom
Latin America :
Caribbean
Antigua and Barbuda,
Leeward Islands
Aruba
Bahamas
Barbados
Bermuda
Cayman Islands
28 December 2007
Page 49
Cuba
Dominica
Dominican Republic
Grenada, Windward
Islands
Haiti
Jamaica
Netherlands Antilles
Puerto Rico
Saint Kitts and Nevis,
Leeward Islands
Saint Lucia
Trinidad and Tobago
Virgin Islands, Us
Latin America : Central
America
Belize
Costa Rica
El Salvador
Guatemala
Honduras
Mexico
Nicaragua
Panama
Latin America : Lower
South America
Argentina
Brazil
Chile
Paraguay
Uruguay
Latin America : Upper
South America
Bolivia
Colombia
Ecuador
Guyana
Peru
Suriname
North America
United States of America
Canada
Southwest Pacific
American Samoa
Australia
Bulgaria
Fiji
French Polynesia
Guam
Kiribati
Marshall Islands
Micronesia Federated
States of
New Caledonia
New Zealand
Palau
Papua New Guinea
Samoa
Solomon Islands
Tonga